Re: Building Big Ass Linux Machine, what are the limits?

ketil@ii.uib.no
28 Sep 1998 09:15:16 +0200


Rik van Riel <H.H.vanRiel@phys.uu.nl> writes:

> With 150+ simultaneous users, you _will_ need some CPU power!
> Go for a dual or quad CPU machine...

I'd suggest SMP too, since you're going to have a bit of interactive
load. Your users are bound to run braindead software, and with a
multi-CPU config, you can hope that some of the programs will only hose
one of the CPUs.

PPro, while a good CPU is getting old, the Quad board costs $1000, and
the PII is probably cheaper. You should be able to put together a dual
PII at 350MHz for the price of the Quad board alone, it will be as fast,
and have higher memory bandwidth etc for free.

> How about using RAID for more reliability? If you're using
> the machine for really big work, I guess you can pay for
> hardware RAID solutions with hot-swappable drives...

RAID mirroring is fast and reliable, but expensive, RAID striping with
error correction, is cheaper in disk cost, but much slower writing.

(When will we see a mirrored volume acting as a cache for a larger
stripe/ecc volume? HP does something like this, I think)

Last time the cache discussion was brought up - a long time ago, surely,
it was concluded that controllers with cache was evil - or at best, a
waste of money, and that that money was better spent on RAM, which is
usually cheaper and faster, anyway. Has something changed?

> 2.0, almost; the 2.1/2.2 series is definately ready.

Wouldn't you want to run 2.2.x, where x is large enough that most bugs
have been weeded out of 2.2.0? 2.1 is still labeled "experimental", and
there's been enough troubles with the latest kernels that I'd wait a bit
longer for an enterprise-critical system.

~kzm

-- 
If I haven't seen further, it is by standing in the footprints of giants

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/