1

Windows server software versions, vary in how many CPU sockets that they support. So if you have a normal server with 4 CPU sockets. If each CPU is a quad core (belive there is talk of 6 or 12 core) then that system has 16 cores to call on.

Do blade servers pool their cores in a similar fashion or is each blade its own machine.

kingchris
  • 244
  • 4
  • 13
  • Performance would be awful if you pooled the cores. Memory and inter-CPU interrupts would be huge bottlenecks. – David Schwartz Aug 26 '11 at 09:05
  • Well a virtual pooling done by Windows, so I could spawn treads, and not worry about inter machine communications. There are packages out there that do that but they all have to be managed and often require extra API calls from with in my code. – kingchris Aug 28 '11 at 08:01
  • The problem is not that you have to worry about and code in a specific way, the problem is that performance would be awful. The memory and cores would be too far apart to function effectively as a unit. It would be like having your CPU in Asia and your memory in Australia. It would be painfully slow. – David Schwartz Aug 28 '11 at 08:35
  • Oh. I do best fit programming by running an exe 10 000 times with random start params. Then I save/remember the best solution. So I would like to cluster slave blades to master blade. Spawn batches of 100 exes. These exes are read from local storage but executed on slave blade CPUs and slave blade memory. "To dream the impossible ..." – kingchris Aug 29 '11 at 05:43
  • That can be done. That doesn't require core pooling. – David Schwartz Aug 29 '11 at 12:02

2 Answers2

8

No, they don't. Each and every blade is a distinct machine with it's own CPUs/IO ports/memory. They do connect to an (active or passive) backplane (which connects the verious IO options like ethernet etc.) and share the cooling and power equipment.

pfo
  • 5,700
  • 24
  • 36
3

Not today, there was a time when a number of blade vendors were looking at using the blade traces to do something like this but given the huge increase in cores-per-die we'll see this year it's kind of pointless and none of them are really pursuing it anymore.

Chopper3
  • 101,299
  • 9
  • 108
  • 239
  • Thanks. Have to wait for the old Lotto win then purcase a 4 socket 12 core each,box. Just to have the most powerful box on the block. – kingchris Jan 15 '10 at 12:33
  • that wouldn't be the most powerful for very long at **all** - wait for what May brings ;) – Chopper3 Jan 15 '10 at 12:35
  • I once worked with a stack of 4 IBM 3950s, each having 24 cores and 128GB and with the stacking kit inserted into the four machines where visible as single system image with 96 cores and 512GB RAM. – pfo Jan 15 '10 at 13:30
  • Yeah they were made machines, but given we'll have 256 thread single servers this year or early next I think we'll see less and less ingenuity like that in the future. – Chopper3 Jan 15 '10 at 13:34
  • To pfo. Could you ask them to send that IBM box round to my house please. I have a spot of Prolog scheduling crunching to get through. And Chopper3. What trade journals have you been reading that we don't. Cheers – kingchris Jan 16 '10 at 04:12