Massive Parallel computing

[Shivkumar]

Dear Sirs;

I was under the impression that superdome supports massive parallel computing and hence they are a kind of supercomputer.

Is the above statement true ?

Thanks,
Shiv

[Joseph Loo]

hi shiv,

based on this:

http://www.hp.com/hpinfo/newsroom/press/2003/031117a.html

and even more

http://www.hp.com/products1/servers/integrity/superdome_high_end/pressroom/index.html

regards.

[Arunvijai_4]

Yes, It is a *kind* of supercomputer with more #cpus and memory. HP Integrity rx2600 Itanium2 Cluster listed #30 in the world.

http://www.top500.org/sublist/System.php?id=6433

-Arun

[morganelan]

Superdome is not a kind of supercomputer.

Superdome is high-end UNIX server with superior:
-Performance
-Partitioning
-Connectivity
-Availability
-Mixing processors

The HP supercomputer is a 1,024-node Cluster Platform 4000 based on HP ProLiant DL145 servers with AMD Opteronâ ¢ processors running the Linux operating system. A high-speed Voltaire InfiniBand interconnect is used with HP's XC System Software providing cluster management capability.

[Mahesh Kumar Malik]

Hi Shiv

Superdome is defenitely powerful computer capable of parallel processing. Following link may be of help

http://www.hp.com/products1/servers/integrity/superdome_high_end/index.html

Regards
Mahesh

[Pete Randall · Accepted solution]

Shiv,

It kind of depends on how you define supercomputer. According to TechWeb's TechEncyclopedia ( http://www.techweb.com/encyclopedia ), a supercomputer is defined as

"The fastest computer available. It is typically used for simulations in petroleum exploration and production, structural analysis, computational fluid dynamics, physics and chemistry, electronic design, nuclear energy research and meteorology. It is also used for realtime animated graphics. See supercomputer sites."

The link to supercomputer sites says

"A compilation of the 500 most powerful computers and their installations was started in 1993 and is updated twice a year. Intended to detect trends and provide reliable tracking of high-performance computing systems, the list uses the Linpack benchmark to evaluate performance and rank the computers. To view the list, which includes system specifications and major application areas, visit www.top500.org."

So, by that definition, a superdome would have to be listed in http://www.top500.org - which it is. At number 252 is a "Integrity Superdome Itanium2 1.5GHz, Hyperplex/640" owned by Cingular Wireless. There are also others.

That makes a superdome a supercomputer.


Pete

[Florian Heigl (new acc)]

For supercomputing purposes You'd have to configure it in a single system image, instead of using multiple npars, so You'll have a maximum configuration of 128 PA-8900 or IPF Dual-core CPUs sided by max. 2TB of memory.

But this is probably where the road gets a bit worse, because while You could cross-connect multiple SuperDomes with InfiniBand, it probably won't scale as well as other systems using i.e. cc:numa, which would scale >512 CPUs, but of course this would depend on the I/O requirements Your application has. And, of course, 128 CPUs are really fast anyway ;)

[Jeff Lightner_1]

Of course its a "Super" computer - its right in the name. I hear they ship with a big S on the front and a red cape. :-)

[Raj D.]

Hi Shiv,

The HP Superdome (Integrity) offers high-end multi-OS capability that allows running all of the leading operating systems â HP-UX 11i, Linux, Windows Server 2003 and OpenVMS (in 2005) â on the same system, even concurrently in a consolidated environment. And now, the 128-way HP Integrity Superdome, with 1 terabyte of memory for running larger enterprise applications and with the cell board design/sx1000 Chipset, is available with the HP developed mx2 module â enabling the deployment of more Intel Itanium processors.

Cheers,
Raj.

[Florian Heigl (new acc)]

Raj, do You know why there's only 16cpu's per partition supported on linux???

This is quite surprising to me :)

[Raj D.]

hi Florian ..,

RHEL3 and Suse9 Supports on Superdome , but I am not sure about the cpu issue.

I think its linux issue , that do not allow more than 16 cpu in a partition.

hth,
Raj.

[Ted Buis]

Definitions differ, but most massively parrallel computing systems have thousands of processors. Designs differ, but most are constructed of many small independent servers that are coupled with a high speed interconnect (e.g. Myrinet) with special software to each handle a small part of the problem. Superdome is much different than these systems in that all resources can be used by a single OS instance or more with the use of nPars and vPars. You could us HP Hyperfabric or Infiniband to build a massively parallel system, but most are built with two processors pizza boxes, like the rx2620 or rx1620 and LINUX and very expensive interconnects.