1 |
On 4/13/06, Brady Catherman <bradyc@××××××.edu> wrote: |
2 |
> |
3 |
> For our Mac OS run I was able to work with Apple to get performance |
4 |
> numbers up. Mac OS started much, much slower than Linux and failed to |
5 |
> scale at all. One of the Apple engineers emailed with several |
6 |
> recommendations to improve performance. The biggest was to not use |
7 |
> Mpich 1. Once I compiled against Mpich 2 everything went much faster |
8 |
> on the Apple side of things. |
9 |
|
10 |
|
11 |
There's one thing I had better try out then, we were just talking about |
12 |
potentially upgrading recently and decided against it as the Myrinet-GM |
13 |
implementation of Mpich was built against 1.2.6. |
14 |
|
15 |
Anyways, the Apple run of HPL was compiled against the Accelerate |
16 |
> framework while the Gentoo side was compiled against the most recent |
17 |
> Goto libraries (in January 2006). The Apple side was compiled with a |
18 |
> custom version of GCC 4.0.2 (XCode doesn't come with fortran) and the |
19 |
> Linux side was compiled with GCC 4.0.2 from Gentoo's ebuild. |
20 |
> |
21 |
> I may have made it harder for myself in that I compiled a 64 bit |
22 |
> kernel and 64 bit platform. |
23 |
> |
24 |
> Our setup is XServe G5's 2.3GHz with 4GB of RAM, running form a local |
25 |
> 80GB SATA hard drive. At the time we just had a simple GigE |
26 |
> connection between nodes. Not we have a GigE based FNN. |
27 |
|
28 |
|
29 |
Same basic setup here except the majority of our G5's are 2Ghz with 2G of |
30 |
RAM. |
31 |
We used the IBM xlf compiler and Goto's libraries in OS X and development |
32 |
ATLAS 3.7.11 (development for G5 support), gcc is version 3.4.4 in Gentoo |
33 |
and Debian. |
34 |
|
35 |
The Gentoo image is true 64-bit whereas the Debian is 64-bit kernel and |
36 |
32-bit |
37 |
userland. Gentoo ends up with a tiny advantage here so far, but we've only |
38 |
tested across 8 nodes, so I wouldn't really count the results yet. |
39 |
|
40 |
Hopefully we'll get our new netbooter up and working soon and I'll run some |
41 |
more comparisons across a much larger set of nodes. |
42 |
|
43 |
Thanks, |
44 |
|
45 |
Justin Bronder |
46 |
University of Maine |
47 |
ACRL |
48 |
|
49 |
with gcc |
50 |
|
51 |
> What environment was your partner using? |
52 |
> |
53 |
> |