Sun on Monday began selling its highest-end Unix servers with new 1.5GHz UltraSPARC IV+ chips. Sun used its milestones and monster servers as fodder for bashing IBM, which recently began a Unix server upgrade. IBM has slotted the Power5+ chips into low-end systems but won’t give higher-end servers the same boost until 2006. Even then, IBM isn’t expected to have a system comparable in size to the E25K.
There is more info here:
http://www.eweek.com/article2/0,1895,1871942,00.asp
The sun rises. Good for them, still plenty of room for big iron out there, like to see one in every garage.
Cause I think a p595 will obliterate an e25k, now. Not next year. Now the e25k may take up a lot more room, I’ll grant that.
Yeah, what’s telling here is that Sun needs a 72-way monster to go after IBMs 64-way.
From TFA : “Analysts, however, would be quick to point out that IBM has outperformed Sun in both raw processor performance and Unix server sales over the past two years. The Power4 and Power5 processors helped IBM gain ground against both Sun and HP.”
From the article:
As with the previous systems upgraded with the new processor, Sun’s CPU/Memory Uniboard and the binary compatibility of the company’s Unix operating system, Solaris 10, will enable customers already running servers with UltraSPARC III or IV to switch out the Uniboards with few, if any, changes to the high-end servers themselves, said David Yen, executive vice president of Sun’s Scalable Systems Group.
It also enables systems running UltraSPARC III, IV and IV+ to operate in the same chassis.
“Our goal was to minimize, if not eliminate, any disruption that needed to be created in an upgrade,” Yen said. “By taking advantage of the hot-plug compatibility, [customers] can do an upgrade on the fly.”
Log in to the server, down a CPU card, let the server transition the processes off of it, yank out the old UltraSparc III board, slap in the IV+, and bring the board back up.
Rolling upgrades on a production server with no down time!
(No, I wouldn’t do it either, but the idea is pretty nifty!)
> Yeah, what’s telling here is that Sun needs a 72-way monster to go after IBMs 64-way.
yup – and the ’72 way’ monster here are really 144 cores! this will really ‘scream’, as in agony when the oracle license prices are to be calculated! and the 64 core IBM p595 will propably still whup it!
> yup – and the ’72 way’ monster here are really 144 cores! this will really ‘scream’, as in agony when the oracle license prices are to be calculated! and the 64 core IBM p595 will propably still whup it!
Oh, I don’t think so. First off Sun and Oracle are close partners and if we’re talking about buying big iron to host an Oracle install, Oracle always works with Sun to at least match the IBM’s price just to keep IBM out of the picture (Oracle hates IBM). I’ve seen this happenning on more than one occasion, so please rest your fears on that one.
As for IBM’s claims that p595 is faster the latest USIV+ SF25K, well, it is a topic for a very long discussion. First of all what benchmarks are we talking about? Sun hardware for instance whips Power5 on Manugistics benchmarks like there is no tomorrow even on the same CPU counts. Same thing goes for data warehousing and TPC-H on very large databases. IBM would like you to believe that they’ve got something phenomenal with Power5, but pretty much all of their benchmarks or so distorted, doctored and unrealistic that they are pretty much useless in real life. Ask IBM why they like to use DDR2 RAM on their benchmarks and hide the real price of the setup. I’ll tell you why, because DDR2 memory easily doubles the price of the system and if put in context that makes IBM absolutely uncompetitive on price. Also ask IBM whey like to turn off all cores but one on the MCM modules for the benchmarks that favour single thread performance and large caches. Bottom line of all this IBM Power5 benchmarks are mostly lies and have little applications in real world, so big time caveat emptor when shopping for high end servers from IBM.
Also in defense for Sun servers, Sun Fire hardware is generally *much* better value than IBM pSeries simply because the lifespan of Sun systems is so much longer. IBM forces you to completely change systems every 2 to 3 years, Sun on the other hand does it approximately every 5 years. So Sun systems have about 2 more years in longivity over IBM, which accounts to some major dollars. With Sun on the other hands you just stick in latest and the greatest uniboards alongside the old ones and that gives you the ability to run the fastest CPU’s in the older chassis, thus significantly stretching the life of the server. I’ve said it before and I’ll say it again — Sun servers are just better.
“Sun hardware for instance whips Power5 on Manugistics benchmarks like there is no tomorrow even on the same CPU counts.”
Let me clarify the subtle difference that Sun introduces with the use of cpu and way:
1 US IV+ cpu = 1 physical chip = 2 cores
1 Power5 cpu = 1/2 physical chip = 1 core
So if we look at the Manugistics Fulfillment v7 benchmark that Sun keeps touting. At
http://www.sun.com/servers/midrange/sunfire_e2900/benchmarks.html#S…
they write:
“Sun’s UltraSPARC IV+ processor-based Sun Fire E20K system running Solaris 10 Operating System (OS) has set a new Manugistics Fulfillment v7 benchmark world record benchmark. The Sun Fire E20K system achieved 66,744,558 SKUs/hr. Sun demonstrates industry-leading performance and price/performance on the 36-way Sun Fire E20K server (9) by beating the IBM P5-590 by 74 percent on performance, with a 2.2x percent price/performance advantage based on hardware category pricing.”
and the (9) refers to:
“(9) Manugistics Fulfillment v7 Sun Fire E20K+ 66,744,558 SKUs/hr, 36 1.5 GHz US-IV+, Oracle 9iR2. IBM p5-590, 38,419,200 SKUs/hr, 32 1.65GHz POWER5. $/Perf calculated using discounted list prices (IBM DDR2). More info http://www.manugistics.com.”
So they are comparing a their 36-way box to a 32-way box. Sounds fair, but if you take a second to translate the units:
36-way = 36 physical chips = 72 cores
32-way = 16 physical chips = 32 cores
So Sun was using 125% more cores for a 74% performance advantage.
The IBM result was about a year ago:
http://www.manugistics.com/news/press_detail.aspx?Id=553
Prices, especially memory, change in a year.
> Prices, especially memory, change in a year.
IBM DDR2 memory for pSeries? Oh no, they wont! Remember we’re talking IBM here and if know anything about their hardware and the way they do business, the memory prices are likely to fall more or less dramatically only with the next major refresh, Power6 perhaps? So DDR2 memory will most likely cost you an arm and a leg on Power5 systems for years to come.
> 1 US IV+ cpu = 1 physical chip = 2 cores
> 1 Power5 cpu = 1/2 physical chip = 1 core
Who gives a crap about a number of cores on the chip? Bottom line of that Manugistics benchmarks is that Sun’s cheaper systems whiped the crap out IBM’s more expensive boxes. Real dollars per amount of performance is the real determinant here, not small time technicalities like core counts, etc.
“IBM DDR2 memory for pSeries? Oh no, they wont! Remember we’re talking IBM here and if know anything about their hardware and the way they do business, the memory prices are likely to fall more or less dramatically only with the next major refresh, Power6 perhaps? So DDR2 memory will most likely cost you an arm and a leg on Power5 systems for years to come.”
Agreed, but this behavior isn’t any different from any other vendor, including Sun. Singling out memory probably wasn’t the right thing to do. It is only 1 of the many components of the cost of a boxes as big as these.
“Who gives a crap about a number of cores on the chip?”
You brought it up with:
“…even on the same CPU counts.”
I simply corrected you.
“Bottom line of that Manugistics benchmarks is that Sun’s cheaper systems whiped the crap out IBM’s more expensive boxes. Real dollars per amount of performance is the real determinant here, not small time technicalities like core counts, etc.”
I’ll admit that this is the first time that I’ve tried to compare Manugistics results. Where are the official reports? All I can find are press releases. The only references that I find with respect to price:performance are the Sun announcement of 2.2 times IBM, and another Sun announcement of 45% better. As far as I’m concerned, all of the price:performance info. is marketing hearsay until I see the configurations in an official report.
“So Sun was using 125% more cores for a 74% performance advantage.”
100% completely irrelevant!
The ‘core counters’ dismiss the much more important facts about total power consumption and performance/watt. Who gives a f@#$ if Sun uses more cores, when each UltraSPARC core is smaller than a single Power5 core. Solaris can handle all those cores!
Some more thoughts somewhat commenting on some of same things as in this thread:
http://www.itjungle.com/breaking/bn101805-story01.html
> Who gives a f@#$ if Sun uses more cores, when each UltraSPARC core is smaller than a single Power5 core. Solaris can handle all those cores!
I’ll tell you who: veritas, SAS, oracle and a lot of other verdors that will love these 144 core boxes… What about suncluster licenses?
Hahah Anonymous, you have really lost it. Stop trying to justify your arguments. You’re only making a fool of yourself.
Number of cores doesn’t matter ? You running OSS on your Sun box or what ? If so, then it doesn’t matter, but when it comes down to packaged apps, it does.
Well, it’s pretty rare that you run a 144 core super
database. Oracle allows licensing based on partitioning
technologies now which means you can just slice that
big server up into more manageable chunks (or cheaper
chunks) using DSD’s and/or zones. IBM also has the
ability to do this with upars on AIX 5.3 but can incur
upto a 40% performance disadvantage and still not provide
the level of component isolation present in DSD’s.
Even USIV+ cannot compete on a pure single thread
performance with P5 – however, USIV+ systems are very
competetive when compared, particularly when considering
investment protection and TCO.