Sun-Oracle x86 server combo tops the SAP charts Sun Microsystems is the quietest server maker in the world these days, but its PR machine - which just pumps out the occasional press release like clues for those trying to figure out its plans since the Oracle gag went on in April - said today that Sun's recent top-end x64 server has smoked the competition in a popular …
" The difference between one and two seconds can nearly double throughput (which makes sense when you think about it)."
Uhm no - if the benchmark is identical, you will normally get slightly more transactions through with a 2 second respomnse time vs 1 second. Of course the penalty of the longer response time is felt by the user, and the concurrency in the server is doubled for the same transaction throughput level. Of coursethe increased concurrency introduces extra overheads of its own, so there is often relatively little improvement in throughput by doubling allowed response time, but it might just enable higher disk & CPU utilisations to be tolerated.
However, the imposition of tighter constraints on response tiomes in benchmarks is to be welcomed, as increasingly complex applications (and integration of apps) makes this essential to maintain user service levels.
We have come to the conclusion and everyone we spoke to at Oracle world that SPARC and Itanium are dead. This leaves Power and x86 which does not include AMD in our purchasing. Our purchasing guidelines are pretty simple:
1) All systems at the end of lease are to be replaced with x86 or Power
2) Develop on Linux and prefer x86
3) If the production workload is greater than four cores put it on Power
4) Put all databases and critical middleware (websphere) on Power
5) We gave all the people in purchasing two coffee cups: and IBM cup for when the Oracle rep stops by and and Oracle cup for the IBMers
Invest: x86(intel) and Power
Divest: SPARC and PARISC (fortunately we never went to Itanium)
Benchmarks are nice....but does not change the market momentum that sparc is dead....but I think Sun knows that and why they sold out to a software company for pennies on the dollar
"6 months later for x64/Solaris"
What utter nonsense. You must have taken MB's FUD class? Throw out lies and innuendo, and hope no one calls you on it.
Solaris.Next is actually being developed on x86 (Sparc just became available)... All fixes must go into Solaris.Next first, therefore, all fixes go on x86 first. If you can't give specifics, then you're lying.
Gaius could be referring to Oracle patch sets rather than Solaris patches but if that's true (I'm not a DBA so feck knows) it won't be true any more.
When Larry is saying Solaris is Oracle's preffered operating system again they could hardly throw out Solaris x86 patches later than Soalris Sparc or Linux and still expect to be taken seriously.
Bet the developers in Oracle are welcoming having dtrace back again!
I doubt that IBM or HP are going to run any more benchmarks on the current kit as they won't compete on a per socket basis.
IBM is will probably publish a whole host of P7 benchmarks when released and HP a whole host when the 4 core version of Itainium is out. HP especially must be hurting due to the delays on the 4 core Tukwila. X86 is just killing Itainium
whereas Solaris uses 256GB RAM, is because then the HP machine can use faster PC2-6400 memory. If HP wanted 256GB RAM, then they could have only used slower RAM. Therefore HP opted for faster RAM, but smaller size. HP also uses 2.8GHz CPUs.
http://www.c0t0d0s0.org/archives/6194-Some-thoughts-about-the-SAP-8-socket-benchmark-at-HP-and-Sun.html
HP is 7 unit machine. Sun is 4 unit machine.
SUN uses slower PC2-5300 RAM sticks and 2.6GHz CPUs and is still 25% faster on SAP. That is because the cores have 99% utilization under Solaris. On HPs Linux machine, the cores are only 87% utiliized, which shows that Linux has problems handling many cores well, on a single machine. On large clusters (which is basically a network with several computers), Linux performs well.
SAP white papers
Linux:
http://download.sap.com/download.epd?context=40E2D9D5E00EEF7CCDB0588464276DE2F0B2EC7F6C1CB666ECFCA652F4AD1B4C
Solaris:
http://download.sap.com/download.epd?context=B1FEF26EB0CC34664FC7E80B933FCCAC80DD88CBFAF48C8D126FB65D80D09E988311DE75E0922A14
If you have enough memory to cache everything, then you will not gain anything from adding more memory. Apparentely, HP saw that 128GB RAM was faster than 256GB RAM. HP has the resources to add another 128GB RAM if they wanted to. And they know how to tweak their system to get the most performance. HP got the fastest result with 128GB ram. Because the memory was faster.
.
There are benches with equal amount of RAM on both Solaris and Linux, and those benches confirms that Solaris machine is faster: Compare SAP benchmarks 2009035 and 2009030 to see that.
.
Regarding Sun T5440 being slower. I have no problems with that. The Niagara is not best on all work loads. If benches show that Niagara is slower, then there is nothing to argue about. Instead, choose SUNs 4 unit which delivers 25% more performance with Solaris, than Linux.
The fun thing is that I dont argue about benches. Whereas there are people here, that refuses to accept benchmarks and white papers. They say "the benches are cherry picked by SUN, or created by SUN to show the strength" etc. The fun thing is, there are so many benches where SUN is fastest. Has SUN created ALL benches? TPC-C, specint, etc? I dont argue about benches. Hard numbers are hard numbers. There are people that wont accept hard numbers. There is one moron that says something like "in MY OPINION the Niagara is slower than Power6, despite benches showing the opposite, I have seen those benches, yes. But in my opinion". Then you can say "in my opinion that 2m guy is shorter than the 1.7m guy" - equally weird.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
