Sun introduced recently the second version of Java Desktop System (JDS) for a flat fee per employee/per year. We tried it and here is what we found out about:
Basically, very few things have changed in the client version of JDS. The desktop and installation looks very much the same as it did a few months ago. The installation has a few bugs: a) some strings haven’t been converted to “Sun JDS” from “SuSE Linux”, b) Evolution and gnome-user-docs packages are stalling during installation because of the misbehaving ‘scrollkeeper-in’ process that takes 99% of the CPU and needs killing to continue (same bug as on JDS1) and c) when trying to edit a user’s groups the font representing the groups is unreadable.
After you’ve installed the system from its three CDs, the Gnome 2.2 desktop comes up (with a few gnome apps upgraded to 2.4 status) and it looks very similar to how it used to look before. Star Office 7-update2 is there, Mozilla 1.4.1 is there, Evolution 1.4.x and Gnomemeeting 0.98 are there too.
And then the big boom: JDS2 has the same bug as JDS 1 and the previous SuSE Linuxes had: some RealTek revisions won’t work because of the patches used on the Sun/SuSE modified kernel. Some Sun engineers emailed me about it the last time I mentioned that, but the problem is just not resolved (the card is fully compatible with Linux and all the other 10-15 Linux distros/versions I have tried it on have worked fine). Inetd is up, the driver module is up properly, but there is no network functionality, not even when trying to ping my neighbor PC through IP. I am pretty sure that if Sun had moved to kernel 2.6 the patches used would be different enough to not trouble the RealTek driver, but it seems that Sun didn’t even bother upgrading their 2.4 kernel above 2.4.19 (same as in JDS1 months ago). Bah.
And then there was the Mozilla resize problem. Now, that is a new bug only on JDS2: resizing the Mozilla 1.4 window to become bigger works great, resizing it to making it smaller it stalls. Other applications don’t have the particular problem, only Mozilla does (however refreshing of any window while resizing is not done in acceptable intervals, giving the feeling that the operation is slow in general).
And then it was the fact that after you left the JDS2 untouched for a few minutes and it would go to the mode of screensaver-locking the monitor would re-sync itself over and over again (making the same mechanical sound as it was changing resolution or refresh rate, but it wasn’t really).
And then there is Nautilus. Sun does a mediocre job automounting all devices/partitions on “This Computer” Nautilus view: someone tell me the logic behind naming a partition as “Disk (3)” (while it is hda2) and when double clicking into it it, it gets renamed as “data 1”. Honestly, that’s confusing as hell.
But there are more problems: for example, navigating to “data1” partition, which is nothing but my Slackware (also kernel 2.4.x and ReiserFS-based), Nautilus would misbehave: click on any file on my personal folder (/data1/home/eugenia/) and they would dissapear from Nautilus! Thankfully the files were not getting deleted (I almost got a heart attack for a second or two), they were just dissapearing from the Nautilus view. Refreshing the folder would show again all files until you simply select a file and it would dissapear again. I don’t remember seeing this bug occuring on JDS1.
As for the sharing options on folders, the UI is terrible (check screenshot). Usability is ranking zero in the 3 dialogs when trying to create a new samba share on a selected folder.
Regarding the games included, the Scores don’t work (they don’t get recorded). There is a caveat on the way you configure the gnome-games source code (both FreeBSD and Slackware has fallen into the trap recently) and Sun’s JDS2 has the same problem. Anyways, that’s a small problem compared to the Stones game which is completely unusable. It gets stalled (unrefreshed windows etc), the music it uses it gets repeated over and over (similar effect as when you run a modern game on a very old PC, however here it is not the case as my PC is fast enough and “top” doesn’t show any high CPU consumption generated from anything) and you have to kill it in order to get rid of it.
Yes, you could argue that having Star Office 7-up2, the brand new Java System Update Service (to update the OS via the net) and the Java Studio IDE worth the $50-100 bucks that Sun will ask you for it. But on the other hand, when you can’t use your browser or your network card on a Linux-certified PC or when Sun has made zero progress to offer new features on the desktop side itself, it is all a moot point. Honestly, JDS2 should have being called JDS 1.1, definetely not 2.0.
The only part of the OS that has seen new features and real progress are the system management tools (coming on the last CD in the box):
Java Desktop System Configuration Manager, Release 1: Policy-based desktop configuration manager. Java Desktop System Configuration Manager defines user settings and provides the ability to lockdown user desktop systems.
Sun Control Station 2.1: Comprehensive toolset for centralized desktop deployment and management . Sun Control Station, 2.1 is ideal for remotely performing desktop provisioning, management, and software updating. The easy-to-use interface also performs dependency checking, health monitoring and asset management.
Remote Desktop Takeover: Remote administration including viewing and interacting with users desktop display to help, guide and troubleshoot. (don’t get excited, it’s just VNC-based, so it would be much slower than X remote display or Terminal Services).
Unfortunately I could not use the new management tools because the server side requires 1 GB of RAM on a machine that needs to run RHEL 2.1 or Red Hat Linux 7.3. I don’t have a machine with 1 GB RAM (except my dual Power Mac which has 2 GB of RAM) and if I had one, it would be on a very recent machine, and a very recent machine wouldn’t guarantee to run with Red Hat 7.3 as this OS is already old. I don’t know what Sun is thinking for not updating this software to work with newer server OSes. In any case, the good people at Sun sent a demo over of the management services running on a Mozilla window that doesn’t require an external server (see screenshot) which worked adequately.
Conclusively whatever problems I encountered with JDS1 (read my review of JDS1) are ALL still here with JDS2 and there are even new ones. I just can’t recommend it, it is not polished and it is not tested properly. I would have expected a better job from Sun.
If the newly announced Red Hat Desktop (part of the Enterprise family of OSes, also targetting the enterprise desktop) manages to be bug-free, stable, faster and offers server side management tools like JDS2 does, I would recommend corporate buyers going with Red Hat. I am not sure if SuSE has something similar in terms of remote management tools, but if they do I would think them as serious contestands too in the enterprise marketplace. Sun’s JDS is just not though. It falls through in many places and it dissapoints on many levels.
Oh, and what’s the deal with the over 200 pages of licensing agreements? Why the need for so much wasted paper?
Installation: 6.5/10
Hardware Support: 7/10
Ease of use: 7.5/10
Features: 7/10
Credibility: 6/10 (stability, bugs, security)
Speed: 6.5/10 (throughput, UI responsiveness, latency)
Overall: 6.75
JDS is just one in a long string of Sun’s software projects that works half-assedly and is brought to market far too early to be worthwhile.
Sun’s various projects such as their “firewall” product that didn’t work reliably enough to work on a 20 client network more than a few hours at a time (and which didn’t support reverse NAT properly) are continued here in the JDS – presold, surely, to clients that will be highly disappointed in what they get from this aging and senile old giant.
And to think I used to be not only a Sun supporter, but a major Sun partner in Canada. Times change, Sun hasn’t.
hahaha… Java Desktop is great, the linux system is just an old version. they simply did not upgrade SuSE. They upgraded the sun software end.
I personally think the Fujitsu + Sun server & CPU line merge is more important
Sun decided to commit to JDS before Novel bought SuSE. I won’t be surprised to see Sun less enthusiastic about the platform these days.
I really don’t see the market for JDS and the project strikes me as mostly a waste of Suns money and efforts.
I would rather see them spend the effort on x86 Solaris to be honest.
whats the waste? if a company with 2 employees puts out puts a bunch of software together and sells it as a linux distro why can’t a multi-billion dollar company?
I do agree that JDS on Solaris woudl be better… Solaris just lacks the drivers that linux has for the desktop.. and then youd get a million articles telling you why JDS on Solaris sucks because it wont work with this and what and this and that…
they could always lock it to SPARC and x86 PCs if they ever decide to enter the PC market like all their competetors are doing..
i forgot to add that the bugs is also a shame. i sent a message to some of the people i know at sun and they reffused to comment because they are embarrassed about it. It must be pretty embarassing.
expect an upgrade to JDS 2 soon.
As I’ve said before, wait till the end of 2005 before making judgements about SUN. They have 18 months to turn the ship around, failure to do so will render thier business useless. They’re failing to generate the revenue and failing to meet market expectations in delievering an alternative to Windows.
Atleast with Novell, they’re realistic and they know what they’re up against, SUN on the other hand is hiding behind their reality distortion field hoping that there will still be enough suckers out there willing to spend $200,000 on an overpriced, underperforming server that makes Itanium look half decent.
I TOTALLY AGREE!
I say if Sun thinks there Solaris OS is the cats pajama’s, make a good supported port to x86. At least then I know they stand behind there product. Leave Linux to Novell/IBM at least they seem to want to inovate on the platform
No I haven’t used it, and I have no desire to do so. Reading about all those bugs, the ancient base (kernel 2.4.19?) and seeing those dreadfully ugly screenshots makes me feel sorry for Sun. I mean, if a billion dollar tech giant can’t technologically compete with one guy and his spare time, on one of their flagship products, you have to wonder what the heck is going on at Sun.
How difficult would it really be to get a base (Redhat/Fedora or Mandrake, it doesn’t matter), slap on kernel 2.6, GNOME 2.6 and the newly GPL’ed YaST? Clear themselves of the SuSE cruft etc. altogether and start building a stable, powerful modern system. While they’re at it, they could make Swing play nice with GTK+2 (I think it does already, or is that just GTK+1?), have all their Java apps use it and have a consistent desktop at the very least.
Currently everything just screams “half-assed”. Not what you’d expect from a company that is supposed to be an industry leader.
Putting my tin-foil hat on here, I wouldn’t be suprised if Sun is trying to offload such a terribly buggy, ugly and outdated GNU/Linux product to give more credence to Solaris. The worst bit is, the pricing isn’t *that* great anyway.
I really like Solaris and Sun’s product but not JDS. I regreted I spent $65 for it. I should buy Xandros it is way
better than SJD. Sorry Sun.
> While they’re at it, they could make Swing play nice with GTK+2 (I think it does already, or is that just GTK+1?), have all their Java apps use it and have a consistent desktop at the very least.
Sun’s version of Java already have this capability. The real problem is the fact that they don’t link all their java apps against the gtk+ engine, so you end up having 5 java apps included in the OS, and all look different:
http://img.osnews.com/img/5286/jds7.png
I pointed this out months ago, and the problem is still there.
>Sun’s version of Java already have this capability. The real
>problem is the fact that they don’t link all their java apps
>against the gtk+ engine, so you end up having 5 java apps
>included in the OS, and all look different:
Filelight is much more impressive than that disk usage app, too. It requires QT, of course. What I’m seeing there looks to be some relatively lightweight applications from Sun’s WBEM packages for Solaris hacked up into Linux duty. They surely haven’t put a great deal of effort into their Java apps, and that’s your reason for not having the GTK+ engine linked in… they weren’t designed originally to run under Linux.
(read modded-down posts for a little meat on Sun’s history)
stand behind solaris x86? a good port? It’s a good port already, solaris 10 will be oustanding. It’s the drivers thats the problem.
I do agree they have until 2005 to turn ship around.
about the product support, sun stands behind solaris.. as you can see since schwartz has become president its turning into a core peice of their business, like java. by 2005 if sun does not become profitable i will believe it is doomed mainly due to mcneely and others trying to live in the past… according to schwartz he claims they know the problems and are applying basic changes and over time they will become profitable.
I’d really want to see the result of the Fujitsu + Sun partial server merge and SPARC development. Will it be like AMD + Fujitsu?
I personally think Sun should buy out PC company like microtel to keep in pace with its competetors.
JDS may not be just a way to make money off linux but since you promoted it as an only alternative you have should burry your head sun, with these false claims it makes sun look low. Maybe its because of all the layoffs……… I think if sun acquired SkyOS and paid their developers it would be better 😉
SkyOS doesn’t do multi-user. Maybe Sun would like that since it wants Solaris on the server… I dunno. But as far as Solaris 10 being outstanding, that’s certainly a matter of opinion (not mine) – and won’t affect JDS much.
The fact that JDS is so buggy, rushed to market and generally based on a Linux distribution (and supporting packages) from the relatively distance past (mid-2003?), it seems that Sun has approached this the way they have their other software…
What software, you ask? It’s mostly gone now, having failed the test of the marketplace, often leaving the sales channel (and consulting affiliates like Solect) floundering trying to figure out how to make their customers happy again (can’t tell you how many dealers had to supply FW-1 free after Sun’s completely botched firewall product hit terminal crapulence… or their IPX-to-Internet integration product bought from Cisco and promptly dropped… or that SMB/IPX/NFS product they put money into, then ruined/dropped… ad infinitum). Take the Raq… please, take it… please?
One thing that leaves me scratching my head is this is Sun: the company that makes Java. They know full well that NPTL support in GLIBC/Linux makes their Java applications considerably faster – in some cases, dramatically faster… and here they are shipping 2.4.19 over a Linuxthreads GLIBC, throwing 30% or better of their Java engine’s performance on Linux out the window. It makes no sense unless you have known Sun for long enough. 1996 was the year they *should* have turned it all around, and only the DotCom bubble has kept them floating on all that exhuberantly wasted cash.
(read modded down posts for some more history if you care)
Seriously, the rest of the desktop looks fine and then all of sudden you start looking at some java apps like the java control center and the fonts looks like something from Amiga circa 1988. And what’s up with Yast2, did they slap a java front end onto the engine and that’s why it looks like dogshit?
Come on Sun, can you assign one engineer to do fonts. It just hurts the eyes to look at it no matter what the L&F is. Swing might have a good api, but until they do something with fonts then I’ll stick with SWT/JFace.
As someone else pointed out, unless they patched whatever 2.4.x kernel they’re using in this, then it’s absolutely ridiculous to not be using a 2.6.x kernel with a NPTL-enabled glibc. It’s common knowledge that Java can get huge performance gains with NPTL.
Really, Sun made a big deal about all the contracts they signed with the chinese for millions of desktops, they rolled out in Britain, and blah, blah, blah. I wonder what it is they are sending out there. Doesn’t sound like their product will do well based on this review. Maybe it’ll damage Linux reputation with those folks for awhile. But hopefully, it’ll just damage Sun’s rep. Shame on Sun for such a shoddy job.
It’s common knowledge that Java can get huge performance gains with NPTL.
Nice to see that you ignore the fact that JDS uses NGPL.
As long as it’s patched for the security holes that popped up around 2.4.23/2.4.24
I think its ok to pick a kernel and basically run with it, to give your developers a stable target.
However, i’d expect to see a complete, and seamless desktop on top of this by now – this is Sun, surely they can build the packages, and do any bugfixing work necessary to integrate them, and then test them properly??
Sacrificing Java performance by not using NTPL/2.6 is an understandable move in my opinion, but what I don’t understand is why every single menu entry and function in a JDS2 bare install doesn’t work perfectly? And why every icon, graphic, toolbar and scrollbar isn’t perfectly matched, tweaked and massaged so it looks absolutely right.
If it cant be made to look and work right, it shouldnt be included.
Sun can’t honestly think that this is acceptable for a corporate desktop?! I think it’s complete sh*t.
Unless they can find some Apple-style aesthetic and quality focus, and some Microsoft ‘give the brain-dead customer a button to click to do everything within reason’, then they’re doomed on the mass-market desktop.
It doesnt matter if its Linux or Solaris powering it, this shoddy workmanship is just unacceptable from a vendor like Sun.
Sun has a big name and it is using to undermine the good faith that other Linux distributors have created for Linux. It is a classic strategy. If enough people come to associate this piss-poor effort of a desktop with Linux, they can then offer to sell them a better version, which will of course run Solaris.
And some poor souls will fall for it.
I have nothing agaisnt Solaris. It’s a terrific server OS. Technically, it is quite good.
I just do not do proprietary OSes that require proprietary hardware to run. I like my Linux just fine. If I need to backup to a big external drive in a hurry, I can just plug-in a large firewire drive and be done. I also have a fairly good assurance that I can install Linux just about on any hardware that I am likely to have around. Not so, with Solaris x86
But getting back on topic, the desktop looks like shit and according to Eugenia, it performs even worse. How is it that they could take a decent product and make it look like crap and perform even worse? How difficult is it to provide a reasonable update to Gnome 2.6 and the latest version of Mozilla and so forth, testing the crap out of the whole thing? Most bugs are fairly shallow and become readily apparent within 2 weeks of intensive testing.
I just do not do proprietary OSes that require proprietary hardware to run.
Then you shouldn’t run any hardware at all. AMD and Intel cpus are proprietary to thier respective companies too.
Most bugs are fairly shallow and become readily apparent within 2 weeks of intensive testing.
Spoken like a person with no experience that just read Catherdral and the Bazaar and took ESRs statements way out of context. You mean to say most shallow bugs become apparent in the first 2 weeks of intense testing, right?
Even that statement is hard to swallow if you have really tested any OS on a box for productizations.
According to a previous post on OSNews, it informs us that SkyOS has gone multi-user.
Java is great, sun is great, Solaris is great, JDS isn’t so great.
it looks like sun paid a part time programmer to slap something together to make a significant profit on it. great job. too bad it will be make the whole company look bad
Reading the review I fail to see where the Java part fits into their “Java Desktop System” thingie. Sure, it’s a Gnome 2.2 desktop with all it has to offer, but where is the Java desktop? It apparently has some IDE for Java development, but is it any better than getting some other less buggy distro for free and then paying for Intelli-J Idea (or whatever the name is nowadays) to get a good Java IDE?
This is the problem in a nutshell. Sun has never really dealt with end-users, their target has always been system administrators and workstation technicians. I am sure that the JDS desktop management tools are really quite impressive-for sys admins. But end-users, people who place more importance on look and feel than on system level management and networking, are unknowns for Sun. Apple used to have the opposite problem-they were the very best OS for end-users but with little or no know-how in the domains of administration and networking-this of course has changed with the introduction of OSX.
People constantly gripe at Linux/FOSS hackers for having no appreciation for the end-users plight-but here we see proof that multi-billion dollar propietary firms are also subject to such weaknesses. Many say that Linux/FOSS software is lacking the raffinesse that propietary money can buy-ie. that full-time paid developers can do all the real work of polishing the look and feel. Well Sun has how many full-time developers working on JDS-and look at what it has brought them.
Sun has made one stategic mistake after the next regarding the Linux market. If they wish to succeed with JDS-they must make a few fundamnetal changes to their approach.
System administrators already expect high-quality from Sun-it is the end-users who need to associate such with Sun and they are failing at this. Sun will never exclusivley focus on end-users-JDS is ultimately geared toward networked installs-the real market of JDS is fat-clients, something which Sun has fought tooth and nail for years.
So they at once need to focus on administrators and flexible seemless integration of JDS into homogenous networks-and they need to focus on end-users at least to the extent that they plan on selling JDS pore-bundled with computers via walmart. In all likelihood Sun is overextending itself and going beyond it’s realm of expertise in pursuing such a broad strategy. They could of course simply decide to pursue JDS and eventually Looking Glass as value-addons for existing Linux products. Their commitment to GNOME is not for JDS but for their own Solaris desktop which badly, badly needed the facelifting. If they choose to continue pursuing JDS as a Linux distribution they need to overcome their resentment of fat-client solutions and by extension end-users. For this scenario-I suggest the following:
Firstly they need to drop JDS’s dependency on the horribly antiquated SuSE versions(Sun appears to be using bits and pieces from SuSE’s server edition 8 and desktop version 8.0/8.1). They should instead focus either on newer versions of the SuSE line-up from Novel or use either debian of fedora as a bais for their own distro. Whatever path they choose they must provide a custome solution for both the desktop and server-they cannot sell a desktop and have large parts of the software repository-ie. their real value-add domain being dependent on a Redhat server. Great if the also support Redhat in this function-but they need to provide this themselves.
If they were to use fedora as a basis they would have to switch to NTPL. Using fedora would give them an “in” on a much bigger community-which would lessen their workload-Sun can’t afford to “go it alone”. A number of JDS developers have mentioned switching to debian at planet.gnome.org-but I doubt that the PHB’s at Sun would go for something like that. If Sun would tie into more recent offerings from Novell they could benefit from the GNOME polishing work being done by the Ximian guys and pioneer implementations of the new udev/HAL led by Robert Love.
They are already working behind the scenes on the newer X extensions-fixes, damages, composite-for their Looking Glass desktop-why not go ahead and commit themselves to Freedesktop.org. They need the support of a larger community to make JDS a success and they need to be able to draw from a large repository of current packages which work with the system without major modification. NGPL is fine if it doesn’t pose fundamental problem to using external packages-it seems although that most of the rest of Linux has chosen NPTL.
Secondly, GNOME makes new releases every six months-Sun should adopt a policy of releasing updates for their system every 6+3 months-ie. 3 months after the offical GNOME release should give Sun the time to do the bug checking and integration work they need for a timely good quality desktop. Moreover JDS could be a showcase for the great things one can do with Java-why in the world Sun hs failed to use this oppurtunity to attract the Java developers it is looking for is hard for me to fathom. I know that latest JDS comes with some great java development tools-but the look and feel of their java applications for JDS or just sad-nothing which promotes their development platform.
Thirdly, Sun must work hard to take avantage of recent kernel improvements-they are selling JDS for consumer desktops and the hardware-support out of the box for new hardware is unparalleled with recent 2.6 series kernels. What they need is a really polished Linux system upon which they can value-add. I will probably never use JDS let alone purchase it-but I would like to see Sun not blow it-this is only make Linux look bad-even if by calling it the “Java Desktop System” there is no reference to Linux upon which the whole thing is based.
Sun’s approach to Linux has always been problematic and self-contradictory. Sun’s ambitions with JDS are just shy of megalomaniac. I am not sure if Sun is even capable of realizing their JDS plans. If they are-they need to start making decisions which are more forward-looking and they need to make commitments to the community. What Sun and Novell are planning to do is not so totally divergent-both wish to create a high-quality Linux product and sell their brand name via value-add on software.
Redhat has decided aginst this route- they are selling service on top of a completely free software. Novell is using mostly free software-but Novell also has a stake in the windows market even if increasingly only from the integration point of view-this makes there propietary add-ons more palatable-and Novell is releasing mor and more of it’s own code under free licenses. Sun is mixing propietary software with a system which is predominately free and selling service.
Its to do with brand recognition. People more readily recognize the Java brand, or so Sun thinks.
Sun doesn’t even really target admins, they target the corporate buyers of large companies.
These guys will never talk about how badly mozilla or other screen saver works. They’ll get some techies to prepare a custom build (in any case, even with a great desktop distro), with the mission to deal with any menial problem.
What they want is the “management utilities” which is a major reducing factor in the TCO equation + support from a large multinational (doesn’t matter if it’s crap) so that they can present something to senior management, with all the management check boxes ticked. And the problems mentionned in the review probably wouldn’t have a check box anyway.
Its to do with brand recognition. People more readily recognize the Java brand, or so Sun thinks.
I wouldn’t want Java to be associated with this. If you’re going to make a Java desktop then you have to make everything Java, or at least provide interface from Java. It is supposed to be a development platform apparently.
It’s the Network Computer and Java Stations all over again – Sun just not understanding what is required, and what is unacceptable.
that would be interesting indeed.
It looks like a distribution from 1995.
Programmers art galore!
> Remote Desktop Takeover: Remote administration including
> viewing and interacting with users desktop display to
> help, guide and troubleshoot. (don’t get excited, it’s
> just VNC-based, so it would be much slower than X remote
> display or Terminal Services).
Remote controling X over VNC is much faster than remote controlling Windows over VNC.
Just finished reading it thoroughly. Excellent in-depth review, good screenshots, points well-made. Thanks for saying things as they are.
Oh, and by the way, Eugenea, you reviews really stink. And I’m talking just about this JDS review, all your reviews are like that. Besides being totally unobjective pretty much all of your reviews are centered around which OS/distribution has got a better looking icons or fonts. Everything is judged by shrinkwrap and not by the functionality of the product.
Perhaps JDS really does stink. Consider that?
I don’t know who would prefer SJD to another linux distro. Most famous distros are a lot better.
I found too funny the buddy xp-imitation icon with the guy with mustach + the white haired woman buddy icon.
Anyway, SUN still hasn’t found a way to make money.
Scrollkeeper isn’t a app that hangs it’s just *insanely*slow*, give it 5 minutes and it will do something.
> Most famous distros are a lot better.
Better in what sense? In actuality JDS can offer better functionality and value for an enterprise desktop than any other Linux distro if you consider the provided remote desktop management functionality (Sun Control Station). Sun Control Station can really make a difference and can certainly outweigh the lack of lustre compared to other distributions. One of the biggest barriers on Linux’s way to enterprise desktop is the lack of management tools to easily manage countless desktops, Control Station addresses that and JDS like it or not is the only Linux variant that can deliver.
Scrollkepper hangs? I don’t believe so…
Scrollkeeper isn’t a app that hangs it’s just *insanely*slow*, give it 5 minutes and it will do something.
And why, oh god, should I give it five minutes of MY time?
Better avoid JDS altogether.
> System administrators already expect high-quality from Sun
No. They did once, but Sun let them down. Harshly. If you aren’t aware of all the stink software that Sun unleashed on its customers, then you aren’t nearly as “on the ball” as you’d like to think. Sun demolished their reputation as a software company years ago. They have Java, and Solaris, and that is *it*. Staroffice is just a modified Openoffice.org, out there for the world to work on. Many of the good things in Solaris are from other companies (Disksuite, Veritas, X, Gnome, CDE, et al). Other than their core software, Sun has nothing high-quality to offer (some would say Solaris is a creaky old mess too… UFS with hacked-in journalling… ugh…). If you work with Solaris and a well-configured Linux system, your eyes will snap open as to how kludgy Solaris is, despite all its vaunted stability.
That JDS stinks of old, patched-together software without polish surprises me absolutely not at all. My Gentoo setup is dramatically more impressive than what I’ve seen from JDS and I’ve never paid a dime for it.
FYI, I have been running NPTL on my Gentoo desktop for months now. It’s been stable, I don’t notice any performance hits in any software, so unless Sun knows something I don’t about a NPTL-enabled OS, they’re throwing Java performance out the window. And, kernel 2.6.7 is shaping up to be *the* Linux kernel… but Sun is using 2.4.19? Are they using EXT2 as their root filesystem too… or maybe EXT3 with its known problems? Feh.
> Sun Control Station can really make a difference and can
> certainly outweigh the lack of lustre compared to other
> distributions.
You can’t be serious. Have you tried to use that kludge?
(read modded-down messages… I was a Sun reseller for years!)
I’d like to see Linux succeed on the desktop.
But no Linux distribution company does, as far as I can see.
SUN has 5-7 billions on the side. It is claiming to be interested to be a major Linux vendor. And yet, it cannot invest 0.5-1 billion to produce a great end user experience (and tie up some server loose ends too). Every company wan’t to have it with Linux on the cheap. Sorry, this ain’t happening.
Sponsor some OS desktop (and I mean with REALLY HEAVY FUNDING)! Pay some artists. Have some usability expert check the final product. Pay to have mp3 licence support and stuff.
On the desktop side it is simple: we want polish, finish, and shine. Stuff has to work together seamlessly. POST INSTALL configuration should be easy.
(Yes, *post install*.
In the zealots pipe dream, ordinary users do not touch their installations, so an easy install should suffice. Heck, they even get Windows pre-installed, don’t they?
In reality a LARGE percentage of users does tinker with their installation. Most people re-install windows for example every year or so. Also, millions of people buy boxed Windows versions to install THEMSELVES, not only a pre-installed PC.
Those people DO tinker with their desktops. Understand that there are not only “my mother” or “grandmother” types out there Linux distros should be targeting to, but millions and millions of people in their 20s, 30s and 40s with a fair bit of computer experience. Those are most likely to be interested in an alternative OS, anyway. Granma wouldn’t even know linux exists).
Hi all,
I recently had an Ethernet problem under SuSE Pro 8.2 (kernel 2.4.20). It turned out that I had to disable ACPI to get Ethernet to work. The overall symptoms were that I’d get hub connectivity (10 or 100 Mb), but I’d get “transmit timeout” messages in dmesg and no IP. Turning off ACPI fixed it all.
Dave
> Staroffice is just a modified Openoffice.org
OpenOffice is a spin off from StarOffice, not vice versa
> Many of the good things in Solaris are from other companies (Disksuite, Veritas, X, Gnome, CDE, et al)
DiskSuite was developed completely in house by Sun and wasn’t brought in from any company. And none of the Veritas products were ever a part of Solaris. Check sources buddy before you post.
> If you work with Solaris and a well-configured Linux system, your eyes will snap open as to how kludgy Solaris is, despite all its vaunted stability.
If you work with Solaris and a Linux system, you eyes will snap out as to to botched and messed up Linux is when you need to patch something almost weekly and forced to move to a next kernel version almost monthly because of hectic and sporadic development surrounding Linux. In many cases you will need some exact kernel revision just to make the damn application run predicatably (Oracle9i anyone?). Solaris is far more stable, predicable, polished, better put together OS than Linux. Solaris is still a hard core Unix without the redhat-config-<blah> or yast utils that make noobs salivate, but nevertheless it is one hell of OS for servers.
I have heard on a couple of occasions that JDS might be moving off of SuSE and on to Debian. Has anyone else heard this?
“Everything is judged by shrinkwrap and not by the functionality of the product.”
Surprise! The functionality of every Linux distribution is exactly the f–king same. Sadly shrinkwrapping is all there is to talk about 🙂
Linux is really just getting boring. I mean how many times can companies or organizations release the same damn thing. Every version of redhat that’s come out since 8 (including fedora) is the same thing. They all look the same, have many of the same quirks (in that there is always something that doesn’t work correctly, thus leaves it completely unusable).
Release cylces are just way to short in the linux world. I used to get excited everytime new distro releases would come around, not any longer. I now have XP on my computer because I finally got sick of loading a new OS on it every month.
Seriously just stop releasing this crap, it’s all the same. Each version that’s released isn’t that much better that it deservers a whole new version. Just release a service pack like Microsoft and apple do. No one should have release cycles any shorter than a year. We’ve seen a lot of compaitibilty problems because of short release cycles.
How is someone supposed to administer a dependable system these days when all people care about is releasing a new verison, and dropping support for the old one as quickly as possible?
To the other extreme you have Debian, their ‘stable’ release is almost becoming unusable even on a server platform. Why can’t a company release every 1 – 1.5 years? Is that too much to ask? Give us a system, let us get it installed, and then support it for a while. Is that such a novel idea? I mean it seems like by the time I would get around to installing the “new” distro another one would come out, and then I’d feel obliged to install that one, despite the fact it’s the exact same damn thing, except it has new version of glibc, gcc, and linux, so it runs .00002% faster. Enough already!
Many of the good things in Solaris are from other companies (Disksuite, Veritas, X, Gnome, CDE, et al). Other than their core software, Sun has nothing high-quality to offer (some would say Solaris is a creaky old mess too… UFS with hacked-in journalling… ugh…). If you work with Solaris and a well-configured Linux system, your eyes will snap open as to how kludgy Solaris is, despite all its vaunted stability.
WRONG. Disk suite is developed by Sun. X is standard on most *nixes. The X that comes with solaris is developed in house, same as CDE. CDE was a joint effort by Sun, IBM and Novell, thus the common desktop environment.
How could you have been an Sun reseller for many years if your basic Unix knowkledge is so lacking? My guess is you are a linux zealot trolling here, and trying to gaing credibility by claiming to have been a Sun reseller. Sorry but your statements speak of your ignorance loud and clear.
UFS in a solaris 9 update gainned tremendous performance gains, ecspecially with logging enabled. Sun just announced the revolutionary 128-bit DFS filesystem. UFS is extremely stable and reliable and now perfroms relatvely well too.
Solaris is kludgy??? Solaris has had technologies for decades that your beloved 2.6 kernel just got recently. Just check slab.c and look at the comment about the slab allocator. NPTL, solaris has default 1:1 thread libraries in solaris 9, enabled as an option on 8. Solaris 10 is light years ahead of linux, linux doesn’t even have a kernel debugger yet kgdb is just being developed. linux’s tracing tools are lacking, look at DTrace. Forget dtrace even the tools like pgrep, pmap, pstack, mpstat kick linux senseless for system observability tools. Linux doesn’t even have crash dump support by default yet. Linux is still growing and the features I have outlined are very important for enterprises.
Go learn something, just becuase you can type configure and make and run linux doesn’t make you l33t hacker who can pass judgement on OS design.
Frankly I am a little disappointed about JDS2 not having fixed even the basic bugs that were published about in reviews. Also Sun’s marketting
However, that being said I am also disappointed that even knowledgeable reviewer seem to hide the fact that they know nothing of the base OS has changed yet write reviews stating the obvious.
Infact, one reviewer even had confirmation from Sun that the base OS has changed.
Eugenia, would it be possible for you to do a review of SUSE desktop in comparison with JDS, since they are both based on the same OS. I would be curious to see if these problems are specific to JDS or anything similar. If they are inherent to JDS, Sun really must do something to fix it. It is regrettable really.
Sun that the base OS has changed.
hasn’t changed
Anonymous: “BTW, I’m writing this from a PC running JDS — it is actually quite a decent product. It is not quite as shiny as other distro’s, but being targetted primarily at the run-of-the-mill desktop in a call center it should do just fine.”
Yes, and I’ve seen other Sun products in call centres: JavaStations. A great way of spending top dollar on dead-end proprietary solutions when bog-standard “no name” PCs could have done the job better and lasted more than the five minutes it took for the JavaStations to become obsolete (and completely unsupported by Sun).
> Yes, and I’ve seen other Sun products in call centres: JavaStations. A great way of spending top dollar on dead-end proprietary solutions when bog-standard “no name” PCs could have done the job better and lasted more than the five minutes it took for the JavaStations to become obsolete (and completely unsupported by Sun).
The parallel you’re trying to draw is absolutely pointless and seriously resembles FUD, since it has no relevance to the product being discussed and carries negativity without factual support. JDS *is* intended to be deployed on “no name” PC’s in attempt to deliver better cost savings and manageability than running Windows or Linux (RedHat, SuSE, etc.) on the same hardware.
Just for everyone to get the facts straight on OpenOffice/StarOffice:
– A German company called StarDivision made StarOffice up to version 4.0
– Then Sun bought it, and released versions 5.0, 5.1 and 5.2
– Then they made some changes, and gave away the sources as OpenOffice.org. StarOffice 6.0 and above are based on that.
— So far the facts —
Now StarOffice was really a nice product up to 5.2. Version 6.0 didn’t have the semi-annoying desktop anymore, but in return it is *much* slower and consumes *much* more memory. Now go figure out who was responsible for that: either the opensource-developers or Sun. As OpenOffice.org 1.0.0 was already as bloated as all later versions were, and as Sun removed the Star Desktop, I also think it was Sun who made it so slow.
That makes Java and OpenOffice about the two slowest applications in the UNIX-world, I would guess. I can only imagine two reasons for that:
– Programmers at Sun can’t code. Unlikely, as Solaris’ base works fine.
– Programmers at Sun all work on those high-end workstations with a dual 1 Ghz UltraSparc III and 2 GB of memory, so they don’t notice how slow it is. More likely, as you do benchmark the kernel (either direct, or indirect by benchmarking a server) but you don’t benchmark desktop apps.
— So far MHO —
adb/kdbg are pretty cryptic programs but they are useful, no doubt. There *is* kernel debugging in Linux, but I’m not going to explain to you how it all works. Go do the research into how the Linux kernel is progressing.
Disksuite was previously a product that needed to be purchased, and became part of the Solaris distribution disks at a later date. Now it’s built into the product. I have Solaris distributions dating back to 1992 in my closet, how ’bout you?
Regarding Solaris 9 and all these fantastic packages that are supposedly without parallel in the Linux world… please elaborate. System partitioning: user mode linux. WBEM: many packages. LVM: needs polish but there. Veritas: available. Your comment about UFS being stable and having decent performance is laughable… it’s definitely mature but it’s not atomic, journalling is stone-age, performance is pathetic compared to XFS/ReiserFS/etc. I have an Ultra 250 here that could not come withing 20% of the performance of four (0+1 RAID) Linux/ReiserFS drives with its 12 (twelve! 0+1 RAID) drives using UFS+logging. I did many real-world tests between the two and the Linux system, which was less than 1/4 the price (Dual Athlon) – the Ultra 250 was utterly stomped… fully patched running Solaris 9 with scalability params tweaked. On the down side, I was not able to get Linux to work properly on the Ultra 250, it had trouble seeing all the drive partitions on reboot; linux’ version of NFS is not perfect either.
It’s obvious you haven’t actually compared the two in place with competent administration. I mean, really, really obvious.
Now, as far as having systems that run forever using the Solaris kernel, I won’t dispute that. It’s why I was a big Sun fan in the mid to late 90s and sold their product, including what was their *FIRST* ill-fated install of a large Javastation network (Sun dropped the ball – period). But you’re a fool if you don’t acknowledge that Solaris moves glacially slow in bringing new features and performance, that it is hamstrung by its ancient/unsupported GNU development utils, that it has archaic package/patch management features, that it has buggy text utilities, etc. Solaris is stable in many senses, no doubt, and for the 24/7 enterprise backroom that can be a good thing – I don’t dispute that; however, that’s a pretty small market that Sun is having trouble keeping above water in.
Ugh. Why do I bother.
That makes Java and OpenOffice about the two slowest applications in the UNIX-world, I would guess. I can only imagine two reasons for that:
Could it be becuase the both offer cross platform GUI toolkits!!!. Anything that uses crossplatfrom tool kits for prtability is always considered slow, mozilla, itunes, wmplayer on MacOs X and on solaris.
BTW, it is the general consensus that StarOffice 6.0 and up are faster on the same hardware/OS than the previous releases. That has been my experience as well. SO7/OO.o 1.1 are faster.
Java is not slow only GUIs developed on Java are. There was a benchmark a few months ago on OSnews on language performance and java could hold a bright candle to any of the other languages in the benchmark.
If something happens to your OS (hacked, maybe?) can you install the OS back onto the RAID, since Disksuite is now built into Solaris? Nope! You boot off CD… and you’d better hope you left some place to put the new OS, because the boot process won’t see the RAIDs. So much for all those recovery partitions you made with 5MB stripes. Just on the off-chance, I called Sun for one of my customers, and they told me there was no way to do it from the CD, fulfilling once again my general impression of software quality @ Sun. Love their graphical manager, tho.
JDS won’t have that problem. Linux RAID partitions can be discovered automatically, as long as the module is in the kernel (or loaded at boot). You can also add an append line at kernel boot if you need to manually specify them for some reason.
>Remote controling X over VNC is much faster than remote controlling Windows over VNC.
Maybe, but it is FAR-FAR slower than Terminal Services, and that’s what matters: what the competition does.
partitioning: user mode linux. WBEM: many packages. LVM: needs polish but there. Veritas: available.
I didn’t mention zones or partitioning in my post, where did you pick that up.
I have an Ultra 250 here that could not come withing 20% of the performance of four (0+1 RAID) Linux/ReiserFS drives with its 12 (twelve! 0+1 RAID) drives using UFS+logging. I did many real-world tests between the two and the Linux system, which was less than 1/4 the price (Dual Athlon) – the Ultra 250 was utterly stomped… fully patched running Solaris 9 with scalability params tweaked.
You are comparing filesystem performance on two completely different systems, one with very old hardware. What are the cpu speeds on the two boxes? You really should learn how to benchmark. put the latest soalris x86 on your dual athlon and run the benchmark.
On the down side, I was not able to get Linux to work properly on the Ultra 250, it had trouble seeing all the drive partitions on reboot; linux’ version of NFS is not perfect either.
Linux has a nastly problem of Data corruption on reboot, with software RAID and SMP configs, go check the linux raid mailing lists. I am not sure what you were running or if you are hitting this particular issue,but just wanted to give you a heads up.
But you’re a fool if you don’t acknowledge that Solaris moves glacially slow in bringing new features and performance, that it is hamstrung by its ancient/unsupported GNU development utils, that it has archaic package/patch management features, that it has buggy text utilities, etc.
Slow in bringing features as opposed to whom, linux, openBSD, freeBSD. Enterprise customers want stability and reliabilty more than new features. But if you ran anything more than a dual cpu box you would know that.
Go look at other Enterprise class OSes and you will see the same slow adoption, low change trend. Notice how redhat, suse use old kernels on thier Enterprise linux distros and how slowly they bring out new releases, relative to thier desktop releases. Notice how Windows Datacenter doesn’t have all the whizz bang features of XP pro. Welcome to the realworld. Sun, HP and IBM Unix divisions know the enterprise market, windows and the linux newcommers don’t, but they are learning. The people who are totally clueless about how enterprises work are linux zealots who just got a taste of what a unix like system looks like and get to compile stuff to make it work. This new experience of power makes them feel L33t and thus think they have credentials to make judgements on market segments they have no clue about. Case in point statements like ” Solaris moves glacially slow in bringing new features”.
Enerprises like the archaic package mangement tools becuase they have spent time over the years to build processes around those tools to manage sytems that run BILLIONS of dollar businesses. They don’t like to change thinks every 6 months. Enterprises are not endusers, Sun makes money selling to enterprises not end-users. So don’t project end-user expectations on Enterprise class OS vendors.
JDS is no different it is targeted at corporate customers not endusers. As I pointed out in an earlier article most of JDS’ direct competitors also have old kernels. That said, I think sun should get the criticism for JDS2 that it is recieveing, they really should aim to be ahead of thier competition not with it.
if something happens to your OS (hacked, maybe?) can you install the OS back onto the RAID, since Disksuite is now built into Solaris? Nope! You boot off CD… and you’d better hope you left some place to put the new OS, because the boot process won’t see the RAIDs. So much for all those recovery partitions you made with 5MB stripes. Just on the off-chance, I called Sun for one of my customers, and they told me there was no way to do it from the CD, fulfilling once again my general impression of software quality @ Sun. Love their graphical manager, tho.
You don’t know how to configure systems and have redundant back up systems, is the jist I got from that little blurb. Yes the issue you raised exists but hit docs.sun.com and read the docs on disksuite on how to save a md configuration to rebuild it later. I hit the same issue as you did and I think is is bad as well but there are documented ways to prevent this from happening.
Enough of this offtopic discussion. Stop trolling please, you have been moderated for being offtopic so stop posting offtpic.
Maybe, but it is FAR-FAR slower than Terminal Services, and that’s what matters: what the competition does.
Yes (against standard VNC) and no.
http://www.tightvnc.com
Use the right one with right settings:) And you might be surprised how good VNC is.
I have used TightVNC and VNC itself on OSX and Linux. It is NO match for Terminal Services’ speed.
We tried to get their Sun One directory server up and running. It ran 5 processes that used about 50 megs of memory each, unshared by the way, and died under any kind of serious parallel load. And their java control software was a joke, it was easier to just crank up ldap_add/modify/delete from the command line as to try and use that thing. We eventually went with OpenLDAP because it just worked, and if we had to write our own tools anyway we might as well use a stable server.
If you get a chance, find the corporate memo on the net where the Sun enterprise developers are taking the Java folks to task for making the worlds crappiest langauge for the Sparc III and explaining why they write so little enterprise software in Jave, but use C instead. Very telling about the way the different divisions at Sun interact.
Regarding benchmarking the two systems together: I took two systems and did a straight task with both of them that used the file system (a slow file population and removal test). One had a CPU advantage (this is a file system test, I/O is supposed to be Sun’s thing, remember) and one had a big I/O advantage – both real and imagined.
Bottom line: the throughput on the Linux box was 500% better in my specific real-world testing (180,000 files, IIRC) at less than 1/4 the price… and that was just counting the base U250 system. The Sun machine didn’t peg the CPU during the testing, the Linux system had high CPU load. All around, it speaks of better design – and I was testing against an early 2.4.x system!! Once i put Linux on the U250 (since it was being obsoleted obviously) I was able to run the same test about 30% faster using only 8 drives (Linux couldn’t see some partitions, remember) as compared with Solaris 9 using 12 drives. Memory consumption on the same machine under Linux was *much* lighter with similar configurations, even at bootup.
Regarding RAID, etc. I have been using RAID on dual cpu Linux machines for some time. I have seen what you may be referring to and had to add an append=”” line to my Lilo configs to work around it. If I was to hazard a guess, Lilo is the one to blame and not Linux, it has never happened to me on a Grub-booted system.
Regarding Disksuite booting – you seem to think that all clients using machines (I am certified up to 32 processors under Solaris and have installed fibrechannel systems using shared chassis, etc.) have the resources you are describing. Not so in this part of the world. Where the client would let us do the job “right” we sacrificed some space on a number of slices to install alternate boot operating systems that we could boot from OpenPROM and recover quickly. Often the customer was in such shock from Sun’s sticker that we were unable to do that for them, or in most cases we came into a legacy S20 system that was being updated to an UltraSPARC chassis of some sort. Your assumption that I don’t know what to do is just that… an assumption.
As for Solaris x86, I’ve used it since 2.4 and once had a fairly reasonable farm of them here. I have one left now, which to its credit is stable enough to keep me from touching the OS itself except to patch it (2.8). However, to test Sx86 against Linux was impossible – there were no drivers in Sx86 for the hardware!
Which brings us back around to Linux and JDS. It may be a substandard distribution, but it’s clear why Sun has gone that route. More than anything, they need the drivers. Sun’s driver subsystem is tiny compared to what Linux has, and Sun’s drivers have always been issues with x86. Back when I had a Sx86 system on one of my servers, the Adaptec controller driver was buggy enough to give me real issues and delayed an install by a few months. When time came to update to Solaris 2.9, I tried Linux (2.4.14 IIRC) and found massive improvements in performance on the same system again. And much more massive improvements in usability, which was hammer nailing the coffin shut on my Solaris infatuation. This is one of several anecdotes that led me to finally conclude that I could no longer remain mired in Sun’s reality distortion field.
JDS stands a chance. Updating the packages it has to something current shouldn’t take Sun long. But you can see from my posts that I have no hope for any of that to happen.
Terminal Services is slower. Or at least RAM consumption for larger ammount of clients is stupendous.
I’ve got bot TightVNC and TS in production use. TS might seem faster on updating smaller parts of the screen, but larger changes drive to higher RAM consumption and to server slowdown.
>Terminal Services is slower.
I disagree. I use XP’s remote desktop from both a Win2k destkop and Mac OS X (MS’ remote software) and it is DAY and NIGHT compared to VNC’s performance.
Diff is the need, yours is very simple:)
1-3 clients TS – very good
more than 3 – well your room gets burning hot. Don’t compare simple needs with production.
For normal work every TS VM needs cca. 40 or more MB RAM, and that is considering light usage. Running some database app makes spawning this app in different VMT and eating memoy again.
2xDual XEON 3GBRAM SCSI RAID5 paracticaly crawls with 15 clients.
Linux and VNC are way not even nearly as slow as this and with 20 clients (btw. machine is the same, app is way more difficult).
I plan to check LTSP just too see where diff is:) Platform or protocol.
Believe me, setup I talk about isn’t household setup:)
Why not just make the IDE part available as a .rpm or .pkg and be done with it. Less work for Sun, who obviously aren’t gonna win any awards for UI design anyway, and better focus…
> I have used TightVNC and VNC itself on OSX and Linux.
Have you tried xf4vnc (http://xf4vnc.sf.net/) or current development tightvnc (i.e. the ones that have an X server plugin)?
Both of them (except that tightvnc is crashing here) show no lag on a LAN, and are pretty usable over a modem.
[i]”Terminal Services is slower. ”
I disagree. I use XP’s remote desktop from both a Win2k destkop and Mac OS X (MS’ remote software) and it is DAY and NIGHT compared to VNC’s performance.[i]
Why not simply grab and XServer session, pipe the bastard through SSH with compression and throw it to a workstation on the other side of the world. Sure, on a 56K connection, it isn’t exactly the most responsive, but over a 10/100 Ethernet, even under heavy conditions, it still performs well.
What are you saying? I already do that: I have my slackware desktop on OSX. On ethernet is fast, but we are talking about VNC here, not about Xnest. And VNC *is* slow even on ethernet.
of them that used the file system (a slow file population and removal test). One had a CPU advantage (this is a file system test, I/O is supposed to be Sun’s thing, remember) and one had a big I/O advantage – both real and imagined.
Well, I don’t know where to begin explaining to you how things work in Oses. You can’t make broad generalizations like I/O is supposed to be Sun’s thing, take an old EOL’d sun box and compare it to a newer box. File systems are an OS subsytem and OSes are made up of instructions, instructions execute on CPUs. The part where the actual DMA occurs is relatively small compared to all the code that needs to execute to setup the DMA. IF you were comparing raw DMA performance, I might give you a little credit. But it looks like you don’t even have the basic knowledge of what systems do. So If you don’t know how things work, how can you possibly benchmark and compare two systems?
Bottom line: the throughput on the Linux box was 500% better in my specific real-world testing (180,000 files, IIRC) at less than 1/4 the price… and that was just counting the base U250 system.
This is a usesless benchmark as I have explained above.
The Sun machine didn’t peg the CPU during the testing, the Linux system had high CPU load. All around, it speaks of better design – and I was testing against an early 2.4.x system!!
Hunh.. Let me get this straight, linux has high cpu utilization on an I/O benchmark, Solaris doesn’t. Linux is designed better because of this fact. Wow I really would love to have what ever you are smoking.
Once i put Linux on the U250 (since it was being obsoleted obviously) I was able to run the same test about 30% faster using only 8 drives (Linux couldn’t see some partitions, remember) as compared with Solaris 9 using 12 drives. Memory consumption on the same machine under Linux was *much* lighter with similar configurations, even at bootup.
First you said you didn’t run the benchmark on linux, now you claim you did, after I pointed out that you are running two different systems!!!
Even if you did run the benchmark on linux running on the UE250. There is damn good explaination for why you might have seen a difference. Sun/Solaris is paranoid about Data Integrity, so much so that it turns out they trun off the SCSI disk’s write cache. They do this in the off chance that a power-outage should occur and the cache hasn’t been flushed, data loss and corruption can occur. Solaris also has a very small filesystem watermark for I/O (this was increased in some solaris 9 update). Solaris makes sure that every write made it to the disk before it returns from the system call. Benchmarks assume a write is complete when the call returns.
Linux on the other hand goes for performance over data integrity. Now let’s see If I was an enterprise and had millions of dollars worth of data, what would I want? Data integrity or a few milliseconds saved on my I/O. The answer is obvious.
Like I said go and learn stuff before you bare your ignorance for all to see.
“Linux on the other hand goes for performance over data integrity. Now let’s see If I was an enterprise and had millions of dollars worth of data, what would I want? Data integrity or a few milliseconds saved on my I/O. The answer is obvious. ”
its the default but easily changed with a mount option in fstab and linux still is faster btw
ts the default but easily changed with a mount option in fstab and linux still is faster btw
You can turn off the SCSI disk’s cache with a mount option in fstab???? what is hdparm for then??? Care to back that up with benchmarks.
> You can’t make broad generalizations like I/O is supposed
> to be Sun’s thing, take an old EOL’d sun box and compare it
> to a newer box
a) I can if I want to. It’s my money.
b) Maybe you didn’t read through sufficiently.
c) You’re misinterpreting the I/O process. It’s mostly a process to copy data from memory directly to the SCSI bus through DMA. But hey, that’s what Sun’s pitch has been to explain their poor processor performance vs. “overall machine throughput”, too.
d) CPU time is very significant. If the Solaris CPU usage was low, but the I/O was slow, what does that tell you? I’ll let you work it out, you claim to be so clueful.
> This is a usesless benchmark as I have explained above.
My actual workload is a useless benchmark as it pertains to how I judge the system’s performance… under my workload? You keep confirming my opinion of your posts, at least.
> Hunh.. Let me get this straight, linux has high cpu
> utilization on an I/O benchmark, Solaris doesn’t. Linux is
> designed better because of this fact.
Like I said, figure it out. You claim to be so smart… why DOES the Linux box work 500% faster using less than half the drives (same drives, FYI) at a file-I/O workload, and chew up significant CPU time doing it, while the Sun box chews up a small amount of CPU time with more than twice the drive I/O potential, and does the work at only 20% the speed? Which system is doing what I ask of it efficiently, and using its resources to improve the speed of my workload?
You probably won’t grok it. No worries, I’m getting the big picture of your admin capabilities by now.
> First you said you didn’t run the benchmark on linux, now
> you claim you did, after I pointed out that you are
> running two different systems!!!
I did, but by that time I wasn’t able to run the machine in the same configuration as the Solaris config. So I don’t think much of the results. Also, ReiserFS is primarily a x86 thing… it’s not stable under SPARC (or other 64 bit systems, AFAIK). I only mention it to let you know that Linux did outperform Solaris on the same machine in very preliminary and partial fooling around.
> Sun/Solaris is paranoid about Data Integrity, so much so
> that it turns out they trun off the SCSI disk’s write
> cache.
I did, too (usually by jumper). But I’m laughing about your “Data Integrity” claim, since UFS is not atomic and AFAIK the journaling on Solaris is stone-age – a simple but clumsy retrofit. Which is why S10 will be shipping with a new file system… guess they aren’t as enamored of it as you are.
Reliability is Solaris’ forte. Up until now, it’s been cost-be-damned, but these days nobody has DotCom budgets.
UFS journaling in recent S9 updates now performs well. So much so it will be the default to have logging turned on in either the next update or definitely in S10.
Journalling under Solaris has improved the speed of some things right from its early days. After some workload testing in Solaris 7, I turned on Journalling and never looked back. It dramatically improves file deletion, in particular.
Morpheus ROOT# showrev
Hostname: morpheus
Hostid: XXXXXXX
Release: 5.9
Kernel architecture: sun4u
Application architecture: sparc
Hardware provider: Sun_Microsystems
Domain: XXXXXX
Kernel version: SunOS 5.9 Generic 112233-11 Dec 2003
Morpheus ROOT#
/ on /dev/md/dsk/d0 read/write/setuid/intr/largefiles/logging/xattr/onerror=panic/dev=1540 000 on Thu Apr 15 12:22:18 2004
c) You’re misinterpreting the I/O process. It’s mostly a process to copy data from memory directly to the SCSI bus through DMA. But hey, that’s what Sun’s pitch has been to explain their poor processor performance vs. “overall machine throughput”, too.
d) CPU time is very significant. If the Solaris CPU usage was low, but the I/O was slow, what does that tell you? I’ll let you work it out, you claim to be so clueful.
WTF am I misinterpreting about the I/O process. You are the one who seems confused.
High CPU utilization for I/O is BAD. The whole point of DMA was to avoid cpu involvement in I/O. Take a guess which system would still run other stuff reasonably well when this benchmark is running, the one with high cpu utilization or the low one.
Anyway your comments are naive at best. I am not a Sys admin. but I am a kernel programmer and have considerable experience in SCSI and IDE systems. That’s why I think your benchmark is bogus and ignorant at best.
You’re an IDE and SCSI programmer, I believe that… uh-huh.
Then you must know that when a system is asked to do a task, it is supposed to do that task with all haste. The fact that Solaris does not, or is not able to, use the CPU to good effect in doing file I/O tasks as fast as possible (the drives can certainly do the task, since the Linux system uses the same drives, and a whole lot less of them) says a lot about the quality of the file system driver, how well the VM works in relation to it, and the quality of the hardware drivers, if not the hardware itself. Since it’s obvious the hardware can do the work far faster than the OS is sending it data, the fact that the CPU is not being used heavily says “this software I/O and/or VM subsystem is poorly designed.” Oh, did I mention that there are two SCSI buses in use on that Solaris box?
The OS’ job is to do tasks as quickly as possible, unless told otherwise using priorities (nice). If it does anything other than that, it smacks loudly of poor design. The OS should *never* assume that it should hold its full capabilities back just in case a new task should come along – it’s the job of the scheduler to either preempt the task/s at the time of execution (if it’s capable of that) or otherwise give up cpu slices to other tasks of a similar or better priority WHEN A NEW AND IMPORTANT ONE COMES ALONG. Secondly, the OS should always flush its write data to drive as quickly as possible, to lower the chance of being interrupted by power, tasks, or software failure.
If you’re such a hot-shit programmer, surely your goals would be the same. But I ain’t buying it, you totally missed what I’ve been saying repeatedly. I think you’re just a squirrel looking for a nut.
Since it’s obvious the hardware can do the work far faster than the OS is sending it data, the fact that the CPU is not being used heavily says “this software I/O and/or VM subsystem is poorly designed.” Oh, did I mention that there are two SCSI buses in use on that Solaris box?
First let me see what you are saying. You said linux running on a dual athalon box is 500% faster on a particular I/O benchmark. You gave no specifications of either the UE250 (cpu speed, SCSI drive which mode, cache or what rpm) or the athalon( cpu frequency, drive rpm cache, SCSI/IDE). You claimed that linux used more cpu on the athalon than Soalris on the UE250.
First give me the specs in detail of both your test machine. I won’t trust any benchmark performed by anyone without all the facts. Why should I just take what you said about the benchmark results.
Ok when I/O is going on, the OS/device driver can only send it new data once the old transaction has been completed. Then the OS sends it new data.
From the looks of it your athalon box (which is newer maybe a 1700+, 3400+, has better drives with more cache and write cache enabled and better controller) hardware performs better than the one on tha old UE250( by the way the UE250 was a 1998 circa machine). You must first give me the full spec of both the machines, honestly.
Also publish the results of your benchmark each and every one of them. Also mention the tuning that was done on both solaris and linux. Please tell me the drive models on the UE250 as well as the athalon box and also the storage controllers. And the scsi drive write cache setting on linux.
The OS’ job is to do tasks as quickly as possible, unless told otherwise using priorities (nice). If it does anything other than that, it smacks loudly of poor design.
According to you Oses only run tasks, Have you heard of interrupts?? nice and priorities don’t do squat for interrupts.
The OS should *never* assume that it should hold its full capabilities back just in case a new task should come along – it’s the job of the scheduler to either preempt the task/s at the time of execution (if it’s capable of that) or otherwise give up cpu slices to other tasks of a similar or better priority WHEN A NEW AND IMPORTANT ONE COMES ALONG.
Soalris does this better than linux, Solaris had a better scheduler than linux and still does, fully preemptitive, O(1) with per cpu run queues and has better locking than linux does.
Secondly, the OS should always flush its write data to drive as quickly as possible, to lower the chance of being interrupted by power, tasks, or software failure.
Flushing the write data on the FS level is different from flushing the write cache on a drive (drives automatically flush thier cache every few milliseconds, unless explicitly done by host software).
f you’re such a hot-shit programmer, surely your goals would be the same. But I ain’t buying it, you totally missed what I’ve been saying repeatedly. I think you’re just a squirrel looking for a nut.
Oooh, sorry to burst your bubble, I get paid the big bucks becuase I know how OSes really work. Your explaination of what Oses must do is not technical enough to warrant a better response.
First publish in full detail the specs and the benchmarks you ran on the two machines, then we can discuss it. Your posts have been confusing at best.
Actually athlon’s didn’t become MP capable till 2001. Nice try. Comparing a Old machine to a newer one and claiming linux is faster. You are trolling.
I’m done with this. Wrapping up.
Athlon boards were MP ages ago, but yes it was a 2001 box.
I don’t have the test data anymore. Nor can I re-run it, both boxes are in production, I re-implemented the SPARC system, which is a dual-cpu machine with 2GB memory and 12 7200 RPM drives (6 internal). Yes the Athlon box is faster @ CPU tasks… much faster… but it had the same drives (7200 RPM 9GB) at the time, now has 15K RPM drives with 66Mhz scsi controllers, and different board, no point in comparing it now even were I disposed towards it.
You go on telling yourself that Solaris’ VM is far better. I’ve seen it for myself, under the exact workload, and I got better utilization of hardware out of Linux. The box kept getting faster as I updated hardware + Linux but it’s now just as old as the Sun box was when I tested it against the Dual Athlon.
As an aside, I’ve never lost any data because of ReiserFS. I cannot say the same about Solaris’ UFS.
I’m done with this. Wrapping up.
Good, I am done discussing with clueless people too.
Athlon boards were MP ages ago, but yes it was a 2001 box.
Double check that. Athlon MP was released in 2001.
I don’t have the test data anymore. Nor can I re-run it, both boxes are in production, I re-implemented the SPARC system, which is a dual-cpu machine with 2GB memory and 12 7200 RPM drives (6 internal). Yes the Athlon box is faster @ CPU tasks… much faster… but it had the same drives (7200 RPM 9GB) at the time, now has 15K RPM drives with 66Mhz scsi controllers, and different board, no point in comparing it now even were I disposed towards it.
Thought so. I know you were speaking out of your ass. Comparing a 3 year old UE250 to a newer Athlon box. You still haven’t provided specs on the drives, how much cache were the drives in the UE250 SCSI II, wide/narrow. Nothing.
You go on telling yourself that Solaris’ VM is far better. I’ve seen it for myself, under the exact workload, and I got better utilization of hardware out of Linux. The box kept getting faster as I updated hardware + Linux but it’s now just as old as the Sun box was when I tested it against the Dual Athlon.
How can you compare OS hardware utilization on two different systems one 3 years older than the other???? I have seen Sun ultra 1s with 167 mhz cpus in 2000 chug along doing (DHCP server and general purpose server at my university) thier thing under decent load and have uptimes on 400+ days.
As an aside, I’ve never lost any data because of ReiserFS. I cannot say the same about Solaris’ UFS.
I have never lost data on UFS with logging either, or even on FAT on windows 95 for that matter. Your point???
Well I could not agree more with article, slow to install
stalls at Ximian Evolution, when it loads after 3rd cd, it does see see Eth0 as working, it sees it, but offers no alternative configuration, the desktop at best is Spartan,
and to add to the bugs if you change your settings and you re-boot you get log in root which is fine then you get
linux:~# and really it gets too confusing, a genuine pain in the but. since I did pay for a license I’ll wait for next OS update and release hoping the Sun Engineers get it together once this year ! Thanks at least I know I am not alone out here, on the good news Linux side, Linspire, is absolutely excellent, almost perfect.. easy to install, recognizes most hardware and connects to your ethernet quickly. CNR is a great straight forward way of getting software.
Double check nothing. MP boards existed before Athlon MP processors. They worked. End of story.
The Athlon is now 3 years old. It’s unbelievably faster now than the Sun box is, and it’s become faster with OS/hardware updates. The machine as obsolete time-wise as the Sun box was, now. Your point is moot, since at the time the best mid-range system Sun had wasn’t much faster than the U250 I had then. Solaris and the high-throughput hardware were positioned directly in the direction of Linux & W2K but were inferior in MY tasks. Say all the derogatory things you wish, you are wrong on more than one level.
And again, you totally missed the point of CPU utilization. It’s because you’re a moron, as far as I can tell. Big-time programmer, my ass. I’ve explained it several times, but you haven’t the skill or inclination to understand. Oh well!
As for reliability, all our Ultra 5 and 10 machines have developed intermittent system bus errors. It’s now a known problem and you can find information about it all over the internet, degraded hardware issues. You can see it particularly well when a RAID partition separates because of a Bus error, and then happens again over a fairly short period of time even though you replaced drive and cable. Older UltraSPARC systems also suffer from occassional memory errors for the same reason. If you were really the bright guy you claim to be, you’d know that too. But you don’t… do you…
As for not losing data with UFS or FAT, I horse-laughed at that. Obviously your level of experience is practically nil. There are known cases where even moderate use of FAT systems will lose data – operating on a few thousand files in a single directory, for example. Get bent, troll.