Slashdot Mirror
Fujitsu SPARC64-GP vs. Sun UltraSPARC II?
AtariDatacenter asks: "I've noticed that Fujitsu is starting to market their
Primepower line of Sun compatible servers, based on the
SPARC64-GP processor. Their Primepower 2000, with 128 processors running at 563mhz, would seem to be a killer of the Sun E10000, which is based on the UltraSPARC II processor. And their server beats the Sun E10k in the top ten TPC-C performance results for non-clustered systems. What do Slashdot readers have to say about the Primepower line of servers?"
I assume this is what became of HAL Computer Systems? Their SPARC64 roadmap looks out of date.
Sorry--it sounds impressive that Fujitsu's hardware won in some TPC-C benchmark, but I've always stood by the saying that there's 3 kinds of lies: "lies, damn lies, and benchmarks." What you need are some numbers from a large real-live application. Something that uses a large database, say Oracle, with several parallel transactions.
I do admit that TPC-C is better than nothing. It's better than Towers of Hanoi or computing primes or timing straight disk I/O.
Does anyone know about the Fujitsu J822s SuperServer? I inherited one with DRS/NX (ICL Unix) installed but I would like to run Linux or a *BSD on it. However, I also don't want to ruin the current installation by blindly experimenting.
OpenSourcerers
You might think that, but you'd probably be wrong. In order to make use of 128 CPUs, they need an OS suitable for the purpose. It's a fair bet that Sun haven't put huge amounts of resources into optimizing Solaris for more than 64 CPUs (sure, they're bound to have played with it in the labs, but it's unlikely to be production ready). Unless you can get something that approximates linear scalability, those extra 64 CPUs are going to be sitting around doing nothing. All this really lets you do is have more domains on a single physical box than an E10K will give you. If they price it right, that alone may make it a success in the crowded machine rooms of the corporate world, but it's not the wonder that it first appears to be. If you want an OS that supports a single OS image across more than 64 CPUs, you can try SGI or DG.
Disclaimer: the above assumes that they're running Solaris, based on the claims of 100% Sun compatibility and 12000 applications that they make on their web site. I guess they could be running a version of SINIX or Reliant Unix that they've optimized for large numbers of processors.
"The invisible and the non-existent look very much alike." -- Delos B. McKown
One thing I can honestly say about Fujitsu, they don't play stupid "100% Sun" games like Sun does -- E.g.,
My local (Orlando, FL) Sun sales/support office is right across the hallway from my company (yes, a whole 20 feet!). I made the grave mistake of going over and asking for a SunSolve CD once. They started asking me for Host IDs, and said that since I had clones (alongside separate, 100% Sun hardware) in the same office, they could not give me one. And what really ticked me in the end was the sales rep trying to sell me a new box! Geez, I just wanted to ask for a SunSolve CD so I would have to avoid downloading a lot of the patches (our Internet bandwidth was limited at the time) and I got my head bit off! Imagine if I really had an issue?!
Fujitsu sells some excellent, re-designed SPARC modules and boxes. In fact, I would argue that if eBay would have gone Fujitsu instead of true Sun, they probably wouldn't have had their "oopses" awhile back with Sun's inadequate airflow causing the ECC cache memories to fail. Their systems are of much better designs, industrial quality and usually cheaper in the end. And, again, their support doesn't bog you down with "100% Fujitsu" questions -- they get right to the problem and help you resolve it.
As such, I'll almost always buy Fujitsu over Sun. And they can (and did) quote me on it (see page 2). Yes, I also support and use x86/Linux, and we are slowly moving away from Solaris as more and more EDA tools are ported to it, but I still need to buy SPARCs on occassion. Fujitsu will continue to get the bill for SPARC.
-- Bryan "TheBS" Smith
-- Bryan "TheBS" Smith
Independent Author, Consultant and Trainer
I'm actually rather disappointed at the discussion so far. I figured I'd see something a little more lively than this. Here's part of what I know:
The SPARC64-GP is significantly faster than the UltraSPARC II at the same clock rates. Yes, the particular web site referencing the SPARC64-GP is old, but the architectual improvements over the USII are the same, and the clock rates have increased.
Of course, the answer is that "UltraSPARC III" is just around the corner. But Fujitsu isn't standing still, and is working on processors and platforms of their own. What I *hope* we're looking at is an Intel/AMD leapfrog kind of competition where we've got some real competition on performance and price.
One area that I've been told that Fujitsu isn't all that hot on (but they are working on the issue) is support/service. They're not quite at the helpdesk level. But I know locally that AMDAHL has been training people to support their new Sun hardware.
Personally, I invited Fujitsu to give our site a presentation next week. I'm anxious to see what kind of things they have to offer. Its apparent that they CURRENTLY have something that can compete with the Sun line in price and performance... from the E220 to the E10k level. I have heard good things about the stability of the solutions that they offer.
I think it is important that Fujitsu starts marketing themselves to Sun shops and giving them an idea of where they are going. I want to see the concentrating on the support side harder. I think I'd be a little scared to run a 64-way box (with domains) connected to EMC via Jaycor fiber cards, and running an Oracle database. If they had a hardcore service division, I wouldn't be so scared.
In the end the E10K is just a toy... Yes, its fast, but I've not heard of a single one that hasn't been split up into domains... Why do you think the V-class was such a failure? Because barely anyone needs a unix system of that size... Performance? cluster... reliability? cluster or mainframe...
and if you anyway go plan on carving up domains, go with a n-class or a couple of 5500s - cheaper and faster...
Yes, but their Ultra 3 is over a year late now, and by the time it is released, it won't even be able to match up with the IBM p-series (RS/6000s) and HP SuperDome in terms of TPC-C, and they still don't have anything coming close to a mainframe (in terms of high availability and performance). Especially with the bad press for the memory corruption problems this past year, they aren't looking good. It's too bad, too - the SPARC architecture is some really great stuff.
Jon Katz - the worlds biggest waste of time and bandwith.
Considering they have been in it for at least 20 years, I'd say awhile. They are migrating away from the S/390 machines to Sun compatible machines.
Since the group which developed the E10k has been purchased by Sun, they have been working on their next generation of the *fire systems, which will be remarkably faster than the 10k. That combined with the new Ultra 3 and 4 processors will probably take the title back within about 6 months
Heck yeah!!
I would get the Sun though, because who know how long Fujitsu will stick around in the server market?