[Beowulf] What services do you run on your cluster nodes?
Perry E. Metzger
perry at piermont.com
Wed Sep 24 06:53:47 PDT 2008
Lawrence Stewart <larry.stewart at sicortex.com> writes:
> I think Greg is talking about HPC interconnects that do OS bypass, and
> Perry is talking about the kernel IP stack. Different things.
True enough. Architectures where the data gets passed to/from userland
directly have different issues. However, you still have to be
scheduled and on processor to do work, and that means you still have
issues if you're on stock PC hardware with stock Linux. (If neither of
those is true, for example in your custom hardware, all bets are
off. Then again, if you're running on custom hardware, the monitoring
issues are totally different, too -- you don't really need to monitor
individual nodes the same way.)
The original question was quite different -- a person with a cluster
of stock PCs was wondering if leaving Postfix on the box to send a
nightly message was going to cause trouble. The answer is, no, it
isn't likely to cause trouble.
> I'm sure the OS kernel is a fine thing. We throw it an occasional
> TLB miss as a bribe not to bother the application. Useful for
> initialization and ECC, but you wouldn't loan it your car.
Your architecture is quite different from what people can build for
themselves, of course, which is a lot of your value proposition.
Perry E. Metzger perry at piermont.com
More information about the Beowulf