BladeSystem - General
cancel
Showing results for 
Search instead for 
Did you mean: 

Itanium versus Nehalem for Java/Cobol highly threaded Workloads?

 
Alzhy
Honored Contributor

Itanium versus Nehalem for Java/Cobol highly threaded Workloads?

Anyone have any info or resources or experiences?

One of my clients specifically asked if a dual socket 4 core Intanium Blade (BL860c i2) will be better performance wise than its Nehalem based cousin BL890c G7 with 2 hexa-core 5660s.

Put simply:
IA 9350 - 8 cores/16 threads (BL860c i2)
HP-UX 11.31

vs

Nehalem 5660 - 12 cores/12 threads (BL490c G7)
RHEL 5.5/6.0

The apps are Java/Cobol highly threaded apps.

And set aside elements like cost, perceieved RAS and TCO.

TIA!
Hakuna Matata.
4 REPLIES
Andrew C Fieldsend
Respected Contributor

Re: Itanium versus Nehalem for Java/Cobol highly threaded Workloads?

Minor point, but Intel's website says that the X5660 is 6 cores and 12 threads, so you would be comparing 8 cores/16 threads Itanium with 12 cores / 24 threads Nehalem.

Having said that, probably the more important questions are around the operating system and compiler implementations (and the JVM for Java). How well do they handle distributing multithreaded workloads across multiple cores and processor threads?
Alzhy
Honored Contributor

Re: Itanium versus Nehalem for Java/Cobol highly threaded Workloads?

Andrew - I am actually pitting dual socket systems - Integrity Blade @ 2CPU,8C,16T versus a Nehlem Blade @ 2CPU,12C,24T.

There are hardly any perf stats for Integrity systems out there considering it's been more than 6 months already since the Integrity Tukwila systems came out.

My "guess" however would be the Tukwila will hold its own and core for core would be "faster".

The Nehalem box would of course be using INTEL Compiler (optimised for the Nehalem chip), JVM is assumed to be optimised as well as do Java. Redhat Linux of course will run the show there.



Hakuna Matata.
Adam Garsha
Valued Contributor

Re: Itanium versus Nehalem for Java/Cobol highly threaded Workloads?

We are in the process of migrating our peoplesoft off of HP-UX comprised of bl860c at app tier and rx7640 at DB tier to RHEL 5.6 with cluster suite/gfs2 at App and DB tiers. For us, it has been a slam dunk win so far. You can get an enclosure full of bl460c G6's with 12-core/96GB per blade vs. much much less memory space on hp-ux front. So cost is much much less for amount of hardware you can through at problem (not even including support costs which are hugely different as well).

Basically, you can get a lot more memory in play for the money, which is huge for java and arguably more important then CPU. We can scale out though so that makes it an easy call for us. Also, we never moved off of HP-UX 11.23 so we were facing an OS migration anyway (arguably).

Other, you'll get 24"cpus" if you enable hyperthreading, assuming you are talking about a 2-socket box with 2 5600 series class chips.

I'd go redhat/x86-64, but that is my 2 cents. So far I'm loving the transition.

Other, we do go straight to redhat for support though too after suffering much pain with keeping contracts up to date via HP and lackluster support (but that could have just been our experiences).
Alzhy
Honored Contributor

Re: Itanium versus Nehalem for Java/Cobol highly threaded Workloads?

AG,

We have been so uber-successful with our UNIX-away Programme too. We have Nehalem-EX's and Dunnigtons and our Migrations are almost complete.

For small UNIX footprints -- we moved them to Linux on a virtual infrastructure (vSPhere or KVM).

And performance and stability has been remarkable so far.

My new client likely has deep pockets that is why I am inclined to recommend the Interity2 Blades... I am just wondering why thee are no benchmarks yet for Integrity Blades versus Proliant Blades -- I know - it's likely a no-brainer for a dual chip vendor to compare the two.

The only RISC platform likely to compete with the Nehalems and the upcoming Bulldozers is likely the Power chip.

Thanks for thine insghts though.
Hakuna Matata.