Re: [SOLVED] Is my processor 32-bit or 64-bit?
On 8/28/2012 5:40 AM, Henrique de Moraes Holschuh wrote:
> On Tue, 28 Aug 2012, Stan Hoeppner wrote:
>> Sadly, due to market realities and diminished customer demand for large
>> monolithic servers, the biggest x86 box Unisys now sells is an 8-way 4U
>> Xeon box. Though with up to 80 cores, 332x times the memory bandwidth,
>> and similarly higher IO bus bandwidth, it runs circles around the
>> monster 32 socket mainframe style boxes of yesterday.
>
> You can go to SGI for suitably large n-way Linux servers. An SGI UV 2k rack
> will go up to 256 Intel E5 processors (2048 cores, 4096 threads), 64TB RAM.
>
> http://www.sgi.com/products/servers/uv/
Do I hear an echo? ;)
I mention the UV and Infinite Storage products here quite often. The UV
and its ancestors, the Altix 3xxx/4xxx IA64 beasts, are a different
class of machine sold into a different market. The SGI boxen are used
exclusively by HPC customers.
The ES7000 and Aviion 25K were used exclusively in traditional
commercial business environments, nearly exclusively hosting databases,
although I did hear of a few large MS Exchange deployments on 16-way
ES7000 boxen.
> I don't think the UV has mainframe-style hardware partitioning. It is a big
> NUMA box with a complex hypercube-like node topology.
The CC-NUMA members of the Altix line can all be partitioned into nodes
via programming the NUMALink routers. Once partitioned, an application
can still span all processors, communicating via MPI and shmem over the
ultra fast NUMALink interconnect. The MIPS/Irix predecessors, the
Origin 2000/3000 could also be partitioned this way, as well as the IA64
Altix 3xxx/4xxx. It is relatively common for customers with very large
SGI NUMA machines to partition them from day one, so a failure on a
single NUMA node doesn't take down the entire system.
This isn't quite the same as traditional mainframe partitioning, as you
can't independently assign memory and IO buses to different nodes the
way you can with many mainframe systems.
--
Stan
Reply to: