| 1 |
Luke-Jr wrote: |
| 2 |
|
| 3 |
> When you have 100 systems, you can probably have the main one do nothing but |
| 4 |
> processing to send it out to the others... Or is that what you did? |
| 5 |
|
| 6 |
Yes, that's what I did. |
| 7 |
|
| 8 |
Let me describe it exactly. |
| 9 |
|
| 10 |
I have two fast boxes. Each is a 2.4 GHz Pentium 4 w/ hyperthreading, |
| 11 |
2 GB RAM. I have one slow box. It's a Celeron 466 MHz w/ 512 MB RAM. |
| 12 |
All run Gentoo, of course. All connected via 100 Mbit ethernet. I |
| 13 |
built on the Celeron, with DISTCC_HOSTS set to "fast1/5 fast2/5". I |
| 14 |
watched performance on all three boxes using gkrellm2. gkrellm showed |
| 15 |
the Celeron 100% CPU bound for most of the build, and showed that the |
| 16 |
fast boxes had very "jaggy" CPU utilization that averaged around |
| 17 |
30-40% during the compilation phase. (By "jaggy" I mean that one |
| 18 |
second would be 90% CPU, the next is 15%, the next is 100%, etc. It |
| 19 |
appears that when the fast box gets a job, it finishes it in a second |
| 20 |
or so and goes back to waiting.) |
| 21 |
|
| 22 |
I guesstimated that the two fast boxes might have roughly 8-10X the |
| 23 |
performance of the slow box. I could be off by a bit. I don't have a |
| 24 |
large fleet of identical boxes to try this on. |
| 25 |
|
| 26 |
-- |
| 27 |
Bob Miller K<bob> |
| 28 |
kbobsoft software consulting |
| 29 |
http://kbobsoft.com kbob@××××××××××.com |
| 30 |
|
| 31 |
-- |
| 32 |
gentoo-dev@g.o mailing list |
Archives