Post a Comment
I found whole interview really stupid, they asked 4 questions and 3 of them was about Microsoft. Whole thing went down in first question where Torvalds said that he doesn't use Microsoft products, sigh why then bother asking other 2. I was so hoping reading good conversation on how Torvalds sees future of, not only Linux kernel, but generally on computer world. Too bad this is another episode of trash journalism.
I too am really disappointed by this "interview". Why oh why do they _always_ have to ask things about Microsoft from Linus?? Can't they just for once leave Microsoft out of the question and ask some interesting and serious questions? Arrrgh! Please, someone do a _good_ interview, pretty please..? :O
Well, "The Computer" IS Microsoft if you like it or not. For example here are latest statistics from my country (Poland). According to them currently about 0,5 % of PC is being connected to the Internet using Linux:
Windows XP 90.8%
Windows Vista 3.8%
Windows 98 2.3%
Windows 2000 1.7%
Linux 0.5%
MacOS X 0.2%
Source: http://www.ranking.pl/index.php?page=Ranks:RanksPage&stat=21|OW
As you can see there are still almost four times more Win2000 desktops connected to the Internet, than Linux machines. And all that is happening in Europe, where awareness of open source is much more widespread compared to US. (see statistics of Firefox adoption)
Whole thing went down in first question where Torvalds said that he doesn't use Microsoft products, sigh why then bother asking other 2.
Because it was an email interview. But I agree that if the interviewer had done their homework, they would have known better than to ask these sorts of questions.
Nevertheless, Linus being Linus, we get very polite, thoughtful answers that refocus the discussion onto the source of Linux's strength and future. I enjoyed reading his responses.
Oh, and that funky, weird language - it's Klingon, dude, definitely Klingon.
Edited 2007-11-26 15:25
"One of the things I personally am really interested in is the move over to SSD [solid-state drives] disks. I'm a huge believer in [reducing] latency, and some of the better SSDs are changing the whole game when it comes to access latency, which in turn has potentially big impacts on the kernel -- and while they are currently expensive enough to be a pretty minor player, that is certainly looking to change in 2008 and later"
Yes, definitely. I agree. I'm planing to buy a new computer with a "small" SSD. It's gonna be strange to switch on your computer, not to hear the disk drive spinning, and GDM popping in a few seconds. But prices have to drop. No way.
"As opposed to what?"
Re-read my statement, please: "but this seems to be specific to PCs, not to computers in general. :-)" There are more computers than PCs, for example, servers or workstations or (near the border of this term's meaning) terminals. Users that grew up with PCs don't find it any strange that they usually are louder than, let's say, a Sun Ultra, a SGI Octane or an Apple Mac Mini - wich are all computers, but no PCs. This is hard semantics, I know, and very individual experiences and understanding of language. :-)
Apples, Suns, and SGIs are PCs too, and Dells have been extremely silent for a long time.
http://en.wikipedia.org/wiki/Personal_computer
Corporations have a lot of money to throw at the sound problem. DIYers the other hand don't have the same resources, so that may be the difference between the experiences.
Two terms
PC = Personal Computer
IBM PC = Personal Computer that is based loosely on the design of the original IBM PC. Or a Personal Computer that is a descendant design of the original IBM PC.
My Silent VIC-20 was a PC as was my ATARI ST. However my fridge computer is not a PC. Just waiting for Apple to come out with a iFridge that is a PC.
Unfortunately repeated readings don't make the meaning any clearer.
How is a Mac Mini not a personal computer?
I can only guess that by "PC" you mean x86-compatible PC. Are you suggesting that it's not possible for a "PC" to run quietly? That's certainly not the case - if you're capable of installing a heatsink and a power supply, about $150 (for a fanless PSU and a low-noise HSF) and a half hour of effort will get you a "PC" that's as silent as any Mac mini.
I had an old Amstrad PC2286, it had a noisy 40mb hard drive, and an even noisier cooling fan. It sort of growled rather than humming that you associate with PC's nowadays.
It was retired after Metallica tried to play in my city, but they were drowned out by my computer.
I'm excited about Solid State Drives as well. For a very, very, very long time, it didn't matter what you did, the hard disk was always a massive bottleneck in any machine, and things haven't really improved that much over many years.
With SSDs, not only do things have the potential to be much faster (with time and development that is), the lack of moving parts and everything that can go wrong with a current hard drive is pretty attractive. It'll also be great if we can just plug them into our current SATA and SCSI interfaces and just carry on. We also get pretty significant power savings on laptops as well! I believe management types call this 'Win win!'. Ughhhh.
As far as storage is concerned, while newer filesystems like ZFS interest me moderately, there's no point in creating a shiny filesystem on top of devices that aren't up to the job. I hope SSDs are the start of creating storage that people can actually reasonably trust.
Me too
I bought an UltraDMA4 SDHC-to-IDE converter on eBay (look for one in a red PCB) and I'm pretty happy with it. I installed XP just for the kicks, FreeBSD is next.
Just beware that XP will REFUSE to run Windows Update with error 0x8DDD0002 if it detects it's installed on a removable drive. Took me 2 weeks to take a wild guess and install the Hitachi driver (actually, filter) that makes Flash-based removable media report itself as fixed.
And boy, does XP boot quick here as well.
I was going to make a SSD using an IDE RAID card and CF-IDE adapters and CF cards but when I did the maths it was way more expensive than a top of the line conventional PATA drive. I could not justify the costs ,the 4GB CF cards I wanted cost about $30AUD+freight each. I have the RAID card and the converters are cheap but the FLASH RAM itself is too pricey.
Does anyone know of a better way to get cheap FLASH RAM. I even thought of chaining USB FLASH drives together but those have similar prices.
So now I am waiting for them to come down in price. Since I plan to build a new computer mid-way through 2008 I will look at one then.
As far as storage is concerned, while newer filesystems like ZFS interest me moderately, there's no point in creating a shiny filesystem on top of devices that aren't up to the job.
ZFS provides end to end data integrity. Check out the "last word in filesystems presentation"... http://www.opensolaris.org/os/community/zfs/docs/zfs_last.pdf
I think thats the presentation that says ZFS loves cheap disks.
Part of the reason ZFS was created was because hard disks are unreliable...so it really doesn't make sense to say that there is no point in creating this filesystem on top of devices that aren't up to the job when that was part of the reasoning behind it.
Actually, it's the opposite... One of the big performance problems with SSDs is the FTL (Firmware Translation Layer) that allows filesystems designed for rotating block devices to manage the completely alien topology of a bare Flash medium (i.e. MTD or Memory Technology Device).
Some of you may know me as a fan of abstractions, but the FTL is a very bad abstraction that does nothing but maintain an elaborate lie so as to defeat any attempts at optimization by the filesystem and I/O layer. It's a stopgap solution at best, and it's never going to exploit the full potential of random-access storage media.
There's no technical reason at all why the OS can't deal directly with the MTD without the evil interference of the FTL. Yes, it requires new Flash-aware filesystems. But our disk-oriented filesystems make absolutely no sense in the context of random-access storage.
Even the "mighty" ZFS is no more suitable to Flash than FAT32. In fact, some FTL implementations will operate more efficiently with FAT32 than with any other filesystem because it's easier to identify free blocks for garbage collection. The more sophisticated the filesystem, the more it looks like a black box to the FTL.
Fortunately, Linux supports raw Flash MTDs today and is developing superior solutions for whenever the storage industry decides to start shipping SSDs without the FTL training wheels. The OLPC XO and the FIC neo1973 are really the first open hardware platforms to feature an MTD. The Asus eeePC and other such UMPCs use an FTL.
Assuming Microsoft is dead serious about getting Windows on the XO, it could only be a matter of time before they push MTD storage support into their mainstream Windows products. Then we might see the storage industry warm to the idea of dropping the FTL.
Today the most commonly used filesystem for Linux MTDs is JFFS2 (Journaling Flash Filesystem), which is actually a log-structured filesystem. Unfortunately, it has some design limitations that make it unwieldy for mass storage media (more than a couple gigs).
The future seems to be LogFS, which, continuing the joke, is actually a journaling filesystem. LogFS has an interesting design that demonstrate a deep level of consideration for the idiosyncrasies of Flash.
As Seen on OSNews:
http://www.osnews.com/story.php/17939
Particularly when it comes to write performance, the FTL implementation can really gum up the works. Some are better than others, they're all proprietary, and most offer no way to bypass the FTL on systems that can support the Flash as an MTD. This is an evil technology that's sapping performance and impeding progress, and we must rise up against it. Down with FTL!
Lots of SSD hype here.
HDD beats SSD in real world desktop performance: http://techreport.com/articles.x/13163/4
HDD operates so quietly it goes unheard in a laptop, and performs so efficiently that it nearly matches the power usage of SSD, making the laptop battery life difference negligible: http://techreport.com/articles.x/13163/14
Large scale reliability is hard to measure, but SSDs and HDDs both have limited life spans.
HDDs are a highly mature technology. They are inexpensive, fast, reliable, quiet, and energy-efficient. They are better in many ways than current SSDs.
So don't buy into the SSD hype too much unless you are really obsessed with that one heavy-load random-read use case that makes flash memory shine.
"They are inexpensive"
Definitely. The current problem of SSD is price per gig.
"fast"
Not really. It takes a long time to load applications and the operating system itself. An ideal disk drive should load the operating system and applications in the blink of an eye. In the real world it can take several seconds to load applications and several minutes to load the operating system. We're used to it but granted this is very slow.
"reliable"
Not at all. This is why we shouldn't expect any disk drive (HDD or SSD) to live more than a few years. Always back up your information. People usually take disk drive reliability for granted and don't backup their data.
"quiet"
Nope. Expecially those 10,000 and 15,000 rpm drives.
"and energy-efficient."
I think SSDs are more energy-efficient, as they are not mecanical.
I think SSDs are better because they are quiet and provide a faster access time. There *is* hype but I think SSDs deserve it.
Not really. It takes a long time to load applications and the operating system itself. An ideal disk drive should load the operating system and applications in the blink of an eye. In the real world it can take several seconds to load applications and several minutes to load the operating system. We're used to it but granted this is very slow.
You forget that it is not just _loading_ things, it's initializing stuff that takes time. If things were so easy that you could just load something in memory and start executing it would be amazing. I measured my HDD performance and it can load data up to ~60mb/s which means that I should be running Linux in about 3-4 seconds. But it doesn't work like that. No matter how fast media you read from the OS still has to initialize itself and all the peripherals.
Nope. Expecially those 10,000 and 15,000 rpm drives.
For Joe User those drives don't matter. Most home PCs have a 5400rpm drive, or a 7200rpm one at most. I have a 7200rpm one and I can't hear it. I can hear the power fan and the GPU fan, though.
I think SSDs are more energy-efficient, as they are not mecanical.
That is, AFAIK, very true. I read somewhere that an SSD drive uses about 1/5 the energy compared to a laptop HDD. Desktop HDDs usually use even a little bit more than those for laptops.
Yep - initialization of the OS is the big time consumer when booting - disk access is not the bottleneck in most cases.
Back in the day, I worked with Honeywell minicomputers. One of the OS choices for those systems was GCOS6 MOD600. We built a configuration for each system and then processed the config, which resulted in a completely initialized OS image stored on disk (and back then, the disks were slow). Booting with the initialized OS image from disk was blazingly fast. Perhaps creating such a capability for Linux would be a cool project.
There are numerous benchmarks showing that there are very few cases where SSDs actually outperform traditional hard disks when it comes to random writes:
http://www.tomshardware.com/2007/08/13/flash_based_hard_drives_come...
That doesn't include the fact that there needs to be a file system which is written which takes into account how an SSD works - to ensure optimum performance and 'levelling'.
Depends upon the drive in question. This review,
from the same site, shows the SSD winning big:
http://www.tomshardware.com/2007/11/21/mtron_ssd_32_gb/
"quiet"
Nope. Expecially those 10,000 and 15,000 rpm drives.
Oh really? 10k and 15k SAS (2.5' drives) are almost silent to my ear. 10k 3.5 disks are not but then, they are mostly uninterresting/irrelevant. 3.5' disks are for bulk storage and not for fast access so 7.2k are speedy enough. For fast access, 2.5' 10k and 15k disks is the way forward. It's all about physics at those speeds.
SSD is about read latency. Sustained read speeds from SSD-disks are lacking or expensive stuff. Also, filesystems and OS:es are very lacking with regards to how SSD works as they do not minimise the number of writes to the disk => bad for life-expectancy of the disk
While I agree that SSD are often over-hyped, there is an issue here in what you're saying: HDD are not both fast and quiet, they are either fast or either quiet.
Also the main drawback about current SSD is their bandwidth limitation for writing, but this doesn't seem to be a 'hard' limit: I expect that in a few years SSD will have write bandwidth equivalent to that of an HDD (by multiplexing communications with different chips), whereas HDD won't have any time soon random-access time equivalent to that of SSD due to the fundamental limitation of the technology.
Well, gigabyte iRam 4gb boots linux and windows on my computer in about 3 seconds. But from a price perspective its not a viable sollution for mass storage. Its neat when compiling code though.
edit: iRam is just that, ramsticks on a board connected through the SATA interface. Not a flash based device.
Edited 2007-11-26 12:56
"gigabyte iRam 4gb boots linux and windows on my computer in about 3 seconds"
Wow...This i-Ram looks pretty cool...
http://www.gigabyte.com.tw/Products/Storage/Products_Overview.aspx?...
http://techreport.com/articles.x/9312
http://www.anandtech.com/storage/showdoc.aspx?i=2480
Question: Is there an i-RAM that has more space? I'm thinking about 10GiB here...
Yes there are I searched until I found this site.
http://www.storagesearch.com/ssd-ram.html
Which lists a lot of companies and products
Here is one that is 16GB @ £1,195
http://www.hyperdrive4.com/
There are larger and larger ones. Ones that fit in rackmount cases or require their own cabinet. StorageSpire takes up to 1TB.
I think the problem is the bigger they are the more costly they are.
You bring up a lot of good points, but in terms of access times an HD which rely on a mechanical subsystem can never beat an SSD which has no mechanical parts.
It couldn't in energy usage either.
-------So don't buy into the SSD hype too much unless you are really obsessed with that one heavy-load random-read use case that makes flash memory shine.-------
Like booting your computer? Some people do that daily, I know I do. I saw a huge difference from the upgrade.
"They are inexpensive"
True
"fast"
Throughput, yes. Latency, hell no.
All my Linux workstation use a 3/4/5 x 250GB SATA in (software) RAID5 array - a >150MB/s sustained read/write setup.
And yet, when I compile my project or start too many VM guests, my soon-to-be-replaced U160/4x36GB/10K/R5 (~70MBps) SCSI setup runs circles around the newer/bigger/better SATA setup.
"reliable"
Not really.
Moving parts == problems. Nothing you can do about it.
New HDs just hide the problems better.
"quiet"
Non-issue.
"energy-efficient."
In the long run SSD will be far more efficient. No moving parts, remember?
SSDs will not replace HDs in the foreseeable future. No doubt about it. SSD are smaller, and far more expensive.
But in the long run, SSDs is the future.
- Gilboa
SSD's are a rather new technology, but the future does look bright. Just take a look at this Fusion IO technology:
http://blogs.zdnet.com/BTL/?p=6394
http://www.fusionio.com/demo.html
Your generalization of HDD vs. SSD doesn't account for the technologies that the SSD are based on. (battery backed) DRAM and Flash SSD have a lot of inherent problems such as limited writes, throughput issues, and cost that will ultimately prevent them from becoming 'HDD Killers'
Now if a technology like NRAM matures, then it will blow the pants off both eletro-mechanical disks, and DRAM/Flash based drives. The theoretical speeds will be similar to that of DRAM except with the distinct advantage of being a non-volatile memory technology.
Anyone who's been though a basic computer architecture course should understand how huge the implications of a technology like this are.
Edited 2007-11-26 16:35
Ideally, if operating systems became cheap enough, it would be great to be able to pull out a chip, and simply drop in a new one in its place - thus, in a desktop, you have two chips, one for storage, one for the operating system itself.
The operating system itself is non-writable, instead all the user files are saved on the user storage device. If that has any problems, load up, format the flash device, and all the settings are recreated on reboot. Updates are handled by storing them on the user writable chip, and the operating system going to the user writable chip before attempting to load the one on the chip itself.
Ideally, if operating systems became cheap enough, it would be great to be able to pull out a chip, and simply drop in a new one in its place - thus, in a desktop, you have two chips, one for storage, one for the operating system itself.
Is "free" cheap enough? -- various Linux distros have been booting from USB flash drives and compact flash cards ever since motherboards and bioses allowed it. And it is simple to just keep all user data and user settings (and even some apps) on a separate USB flash drive or compact flash card.
I was running Puppy Linux from a USB flash drive, and it was lightning fast, because it defaults to loading the entire OS into the ram. It is so small, that I didn't need a separate flash drive for the userspace.
Windows can also boot from a USB flash: http://www.tomshardware.com/2005/09/09/windows_in_your_pocket/
As can OSX (although it seems to require at least a 1-gig drive and seems more complicated to set-up): http://blog.bradbergeron.com/2006/11/29/howto-install-and-boot-os-x...
I'm glad that Linus wants to put more emphasis on this.
I run an SSD, a CF card. But I've been having problems with it at boot time, UDMA timeout errors specifically. Other than that, it's great.
I've got a huge posting about it over at phoronix forums, screenshots too if anybody wants to read about it.
http://www.phoronix.com/forums/showthread.php?t=3424
Edited 2007-11-26 13:03
Whether Linus really feels that was a complete waste of time doing that interview. Whenever Microsoft questions are asked, he should point them to the last 10 interviews he's done with other sites, and tell them to look there for his answers. Seriously, Linus is an extremely intelligent man, so why waste his time, asking him to make comparisons between Linux and Microsoft, when he obviously has no interest in Microsoft. His passion lies within the Linux kernel development.
Ooh, ooh, ooh, here's mine: "If Stever Ballmer asked for your advice, what would you say?"
Ugh...
Seriously, if OSNews does an interview with Linus, there better not be any questions about Microsoft or its products. No boxers or briefs inquiries. Nothing about what features he expects in 2009. Let's try to keep it to questions we can't quickly answer by consulting Jonathan Corbet's Linux Weather Forecast:
http://www.linux-foundation.org/en/Linux_Weather_Forecast
Ask him about his day-to-day work activities. Ask him about code quality and complexity. Ask him about developer relations. Ask him about any long-range concerns he might have about the project. Maybe ask him about license fragmentation between Linux and OpenSolaris and reciprocity with the BSD communities. Ask him about how convergence of logic and stream computation in next-gen hardware might impact the kernel.
People have a huge misunderstanding of the kinds of things that Linus cares about. Ask him about the stuff that matters to him, and you'll have a great interview. Ask him about what matters to the tech news echo chamber, and he'll roll his eyes and phone it in.
I think the fundamental problem of harddisks is that, based on their design, they are very limited in how they can be improved.
Reliability _could_ actually be improved.
Let's just say some manufacturer produces extra small drives and puts three of them plus a raid controller into an ordinary hdd case. This raid 5 setup could appear as a single drive to the customer and would have greatly increased reliability at about twice the price/GB of an ordinary harddisk.
The capacities of hdds will also continue to grow as will their transfer rate which is roughly inversely proportional to the length of one bit on the platter.
What will stay about the same is the latency which basically consists of two parts
a) the time needed to move the head
b) the time needed to rotate the desired zone under the head
This second time is on average equal to half the time needed for one rotation of the platter. This means the average latency is:
T>=0.5*60s/speed_of_hdd[rpm]
For a 7400rmp drive this is about 4ms.
While this doesn't sound like much let's try and put it into proportion:
- It's already about 10% of the time needed to notice that a video/audio-sample or game is lacking
- It's 8,000,000 cycles for a moderately old CPU@2GHz
- It's the read time of a 240kB file @60MB/s
Yep, that's right: If your typical filesize is only a couple of 100kB, the new hdds probably won't be noticably faster - unless they spin faster and therefore run hotter and louder.
Of course, that doesn't mean that people won't use a hell of a lot of lipstick on the pig that hdds are to make it work for another couple of years.
Heck, even tapedrives still got plenty of use cases.
One example is the CERN. They calculated that to accomodate their data rates they'd need their own harddisk factory and since tape is cheaper and they only do sequential reading/writing they opted for that.
For regular desktop use, however, I expect SSDs to dominate the market in the near future.
(I was tempted to say ten years but people would have accused me of using my behind for data mining - and rightly so!)
I've been wanting to buy a hybrid drive for a while now, but I just haven't encountered them in the market. These drives combine flash with a normal platter for a little of the high-speed random-access along with the typical amount of large-scale fast sequential storage. I was really excited when I heard about that technology and I'm pretty disappointed that it's not visible now. Just take 4 GB of flash and slap it onto a HDD and you get the best of all worlds.
They are indeed used by some laptop manufacturers, I've seen them in some Dells and Sonys, but with about 256 MB of Flash memory, there isn't any benchmark that shows even a small increase in performance.
I think SanDisk's Vaulter Disk is going to achieve better results, but I haven't found any benchmarks yet. I wrote about it in my blog (just the announcement, technical details are out of my skill):
http://optimitza.cat/news/2007/11/14/vaulter-disk-the-long-awaited-...
Interview read like a poor effort to get Mr Torvalds to stoke the fire for a flamewar between Linux and Microsoft or grab a few headlines in the media at least.
I think the problem is that what vision people have or would like for Linux is not the same as Mr Torvalds'.
I'd like to see some graphs plotting flash prices against HD prices.
I have a few data points, around 2000 a 256 MB compact flash cost ~$300, or $1200/GB. I just checked a price of a Corsair 16G usb stick for $130, or $8.12/GB. Thats a 147x increase per $. That's faster than moore's law (2^7=128).
Compare HDs. Around the same time, I put in a 40G in my Tivo for at least $150 (probably more). Now, $150 will get you a 750G. That is a 18x increase.
Its pretty obvious that flash capacity/$ is increasing rapidly. The shapes of those two graphs will determine how quickly flash will be cheaper than HDD. It will happen though.
Please elaborate: what was so bad about Ubuntu 7.10 compared to previous versions?
No, actually, I would not.
If Ubuntu 6.06 and 7.04 had been working good, why not reinstall these?
What feelings would those be? Do you really dislike Microsoft, or are you simply parrotting what you think a Linux enthusiast would say?
My experiences with Ubuntu is that everything just works. I've installed it on the PCs of non-techie friends, and I haven't had a single complaint yet.
I'm sorry, but I'm quite skeptical about your whole story. In fact, it sounds like a complete fabrication destined to discourage others to try Linux in general and Ubuntu in particular.
Some weeks ago I put a list of main "bad things" I encounter in latest Ubuntu, not present in previous versions.
A quick reminder:
- broken debmirror (for me it was critical since I use it)
- screensaver settings in LiveCD - activating it the same moment GNOME is displayed - at first sight it looks similar like problems with monitor mode being detected wrong.
- "screen and graphic" xorg.conf's ultimate destroyer.
- graphics card and monitor detection failed on my two PC - 6.06, 6.10 and 7.04 detected them fine.
- Synaptic and Add/remove works horribly with offline repos DVDs. Even created with care - "universe"






