“Sun Microsystems’ product plans are up in the air pending its acquisition by Oracle, but the company’s chip engineers continue to present new designs in the hope they’ll see the light of day. At the Hot Chips conference at Stanford University on Tuesday, Sun presented plans for a security accelerator chip that it said would reduce encryption costs for applications such as VoIP calls and online banking Web sites. The chip, known as a coprocessor, will be included on the same silicon as Rainbow Falls, the code name for the follow-on to Sun’s multithreaded Ultrasparc T2 processor.”
Niagara comes out with built-in SSL accelerator, 3rd party accelerator comes out soon after and turn it into mince meat. T2 comes out with better built-in SSL accelerator, 3rd party accelerator comes out soon after and laughs in its face.
Does anyone see a worthless trend here?
If Sun isn’t careful, they’re going to die… oops, too late…
That really depends on the difference in cost, adds another breakpoint by using a dedicated card and also impacts support contracts.
Yeah you can offload the SSL to dedicated card but if you can get it as a standard part of the chip then that’s a good. If it ends up costing you at extra $5k then that’s probably not going to fly.
So, Sun shouldn’t include SSL acceleration because they eventually get leap-frogged by a third party?
Or, maybe third parties shouldn’t create SSL boards because they eventually get leap-frogged by Sun?
I like the latter better.
You’re always going to bet on the third-party in that scenario because there’s any number of ways in which they can improve performance, and there are already good accelerators around right now. From a commodity perspective it remains to be seen whether Sun’s approach performs better than a x86 system dedicated for the purpose.
It’s an ultra-niche and ultimately fruitless selling point when you consider the competitors that are already there and doing it.
Edited 2009-08-26 23:38 UTC
So it’s a fruitless selling point when there are competitors who base all of their selling on the same point? If you are already buying SPARC, I can’t see how not having to spend any money on a third-party board could be considered a bad thing.
I don’t see how this is any different than a UNIX vendor back in the day bundling a volume manager. It may not be the best one, but it relegates the dedicated third party product (eg Veritas) to the niche, not the other way around.
Same deal with integrated video. I bet integrated video chews up a third of nvidia’s potential market by the mere virtue of being there.
Yes. The phenomenon of trying to push yourself into a niche selling point because of other difficulties you have where there are already existing and established players makes life very, very difficult.
SPARC is a niche market, so it’s hardly going to attract more people to using SPARC which is what Sun need. However, I’d like to see some figures from Sun as to how much business they expect to gain by targetting existing SPARC users who know they need SSL accelerators and already have them installed. It’s easier just to get yourself more powerful hardware if you want to cut costs and can get away with it.
The research, development and production costs versus the potential returns on this seem more and more non-sensical the more you think about it.
You’ve hit the nail on the head with that statement. It’s rather like what a Unix vendor would do about fifteen years ago to justify the cost of their own hardware and software. Unfortunately, with the proliferation of cheap and powerful commodity x86 hardware, cheap storage and freely available storage management subsystems like LVM, software RAID and ZFS, as well as very general purpose storage like Amazon’s S3, that strategy has ultimately become a totally fruitless exercise that has put many big Unix vendors out of business.
When you’re using it as a strategy to try and sell a declining platform the opportunities for success are pretty limited. Sun have always had some very strange ideas about what it is that they’re trying to commoditise. More powerful x86 hardware and cost reasons have already relegated coprocessors and on-chip processing of specific protocols to a niche market themselves. Trying to compete in an already niche market is doomed to failure.
Possibly, but graphics is a large and expanding field in itself with ever expanding requirements that has some way to run yet. You might get total onboard and on-chip graphics at some point in the future, but that is some way off. SSL and encryption/decryption acceleration is certainly not like that so the comparison is rather flawed.
Looking at the presentation at http://blogs.sun.com/sprack/resource/hotChips_spracklen-final.pdf it seems to be an incremental development on what they’ve been doing for the last 5 years or so. In proportion the costs might not have been that bad.
Regardless, I am in favor of this kind of thing (diversity in systems) for the simple reason that if everybody just sits back and says “x86 and Windows / Linux kicks everything’s butt, why bother” then we accept that the state of computing in 2009 is the best of all possible worlds and set in stone from now on. I find that hard to accept. I like cheap x86 as much as the next guy, but I don’t like a monoculture. What if AMD dies? Anybody want to go back to $1000 P3s in desktops, like 10 years ago? I don’t think it’s that far-fetched.
Definitely a double meaning there . But I’d argue that having it bundled adds value beyond the cost of the third party license you save. The reduction in complexity and support finger pointing is worth something.
Anyway you have some valid points. But if Sun fails and systems innovation slows down proportionally, I think it’s a net loss for computing; at best it’s not going to make x86 any cheaper!
Yer. I’m afraid off-loading specific functions to coprocessors in hardware for performance reasons, apart from specific areas (the ever expanding requirements of graphics for one), has never been terribly successful. A dedicated accelerator for such uses is always going to turn out to be better.
I can only see this as a way for Sun to squeeze their Sparc development into ever more niche areas, and niche areas people don’t need at that, before Oracle inevitably has a look at the figures and pulls the plug.
Don’t see the problem. If I can have either a super low volume $2000 board, with third party drivers, versus integrated “dont have to do anything” support in the vendor’s own OS and system, I’ll pick the latter unless the dedicated guy is better by an order of magnitude AND it actually matters.
What I think is also interesting is how Sun has removed UltraSPARC-AT10 related code from OpenSolaris – so one really wonders where Sun’s future is, in regards to hardware design. I simply don’t see Sun being in the SPARC business for much longer as Fujitsu are having less of an interest in that area as well.
Edited 2009-08-27 03:43 UTC
It’s a pretty unambiguously bad sign I’d say. But then again, Sun has been in a bad spot regarding processor design for probably half of the time they’ve been making processors. If it was all about processors, Alpha and MIPS would be alive and SPARC and x86 would be dead already.
I can’t see anybody exiting the high-end Solaris business yet–I’d expect HP to drop HP-UX first. But I find it fascinating how IBM bought Transitive, and then completely buried them after the failed Sun merger (website doesn’t even exist anymore). So maybe the brains at IBM figured they could keep the high end Solaris market without SPARC.
Things started going backwards when their UltraSPARC III delivery started to slip; if it were delivered on time and there was a quick transition to Fujitsu SPARC64 processor – Sun would have remained competitive and profitable. The problem is that that they take too long to make any decisions – decisions that need to be made within days not weeks or months as what happen with Sun. Every decision Sun made seemed to be long, laborious and management never willing to stand up and say, “this is our vision and by hell or high water – if you as an employee are going to undermine it, we’ll boot you out”.
One only needs to look at the forces within Sun who undermined open sourcing Solaris, undermined getting Solaris on x86 as a first class citizen, the undermining of open sourcing Java. If management at Sun had a backbone – those involved would be kicked out.
Solaris has a future without SPARC but it would involve Sun accepting that reality instead of pushing products that aren’t pulling their weight – when I mean pulling their weight, I mean bringing in the cash not only to make a cash profit but to also cover the investment required on all the components that make up the product. Sun would be better off developing high end chipsets and associated hardware and adopting Intel’s processor. There is no reason why Sun couldn’t put together a 32 way Xeon big iron running Solaris with all the same level of reliability as their SPARC hardware but without the massively uncompetitive price tag.
Edited 2009-08-27 06:00 UTC
Well I don’t see why it’s a bad sign, after all Sun business for quite a long time was HW based. True that they should have take less time to make decisions, but again their purpose was all about over-engineering. Still today, costumers have Sun systems running for over 5 years in a row and they worry is if the data center needs electrical maintenance and they go “I won’t be able to show the uptime anymore”.
Anyway, the CMT design is quite good for multi-thread apps check “http://finance.yahoo.com/news/Sun-CMT-Servers-Set-New-World-bw-1913… though for data warehousing it’s best the Sparc64.
Just my though’s
Can you hot-plug CPU/memory boards with Xeon (or Opteron) servers like you can with mid-range and high-end SPARC systems? A lot of customers want this functionality and still buy RISC servers because of that.
Did you read what I wrote or did you just scan what I wrote? “Put one together” means more than just assembling it; it would require implementing features at the chipset level which RISC high end midframes have. I understand that the platform as it exists now is limiting but there is nothing stopping Sun from working with Intel and getting these features implemented.
Edited 2009-08-28 08:42 UTC
Even now, the 1.4GHz Niagara T2 is four times as fast as a 4.7GHz IBM Power6+ on certain benchmarks. For instance Siebel 8.0, were one SUN T5440 with four Niagaras is twice as fast as three IBM Power570 servers with 12 Power6+ CPUs together.
The T3 will have 16 cores, each core able to run 16 threads, which gives 256 threads. The SUN T5440 four Niagara T2 has 256 threads in total. In best case the Niagara T3 will be as fast as one T5440? There are rumours of one 8 socket Niagara T3 box, with 2048 threads. That box will certainly smoke all competition regarding high throughput.
http://www.theregister.co.uk/2009/08/26/sun_rainbowfalls_hotchips/
This will also be a big problem for OSes that try to utilize such many threads if the OS is not scaling well. But Solaris is prepared and handles that well. Granted, Linux scales very well, but that is on large clusters, just a network of computers. I therefore suspect Linux will have big problems running on this machine with one kernel handling many threads.
On the other end, Fujitsu has a 8-core SPARC “Venus” CPU with 128GFlops. It will be extremely fast and use less energy than most other CPUs.
So make your choice: massive throughput? Choose Niagara. Execute few threads very fast? Choose Fujitsu.