1 |
For our Mac OS run I was able to work with Apple to get performance |
2 |
numbers up. Mac OS started much, much slower than Linux and failed to |
3 |
scale at all. One of the Apple engineers emailed with several |
4 |
recommendations to improve performance. The biggest was to not use |
5 |
Mpich 1. Once I compiled against Mpich 2 everything went much faster |
6 |
on the Apple side of things. |
7 |
|
8 |
Anyways, the Apple run of HPL was compiled against the Accelerate |
9 |
framework while the Gentoo side was compiled against the most recent |
10 |
Goto libraries (in January 2006). The Apple side was compiled with a |
11 |
custom version of GCC 4.0.2 (XCode doesn't come with fortran) and the |
12 |
Linux side was compiled with GCC 4.0.2 from Gentoo's ebuild. |
13 |
|
14 |
I may have made it harder for myself in that I compiled a 64 bit |
15 |
kernel and 64 bit platform. |
16 |
|
17 |
Our setup is XServe G5's 2.3GHz with 4GB of RAM, running form a local |
18 |
80GB SATA hard drive. At the time we just had a simple GigE |
19 |
connection between nodes. Not we have a GigE based FNN. |
20 |
|
21 |
What environment was your partner using? |
22 |
|
23 |
|
24 |
|
25 |
|
26 |
On Apr 12, 2006, at 8:26 PM, Justin Bronder wrote: |
27 |
|
28 |
> We've actually noticed the opposite results here with OS X 10.4 vs. |
29 |
> Linux as far as HPL goes. |
30 |
> I don't have the numbers on hand, but Linux was notably faster with |
31 |
> HPL |
32 |
> over both GigE and |
33 |
> Myrinet. In fact the numbers for Colsa (our partner) in the Top500 |
34 |
> were |
35 |
> achieved using Yellow |
36 |
> Dog instead of OS X. That being said if you have a moment I have a |
37 |
> few |
38 |
> questions: |
39 |
> |
40 |
> 1.) What libraries were you using? Did you use Goto's? |
41 |
> 2.) Which compiler(s) did you use? |
42 |
> 3.) What sort of setup do you have? Diskless or not and so on. |
43 |
> |
44 |
> |
45 |
> Brady Catherman wrote: |
46 |
>> Unfortunately BPS has an annoying habit of overwriting the test |
47 |
>> output |
48 |
>> files every time it is run regardless of which test you actually run. |
49 |
>> I didn't notice this when I was doing my initial stuff on RedHat so I |
50 |
>> lost all the other tests results from BPS. I have better tests of the |
51 |
>> hardware since then. We ran HPL on the PowerPC cluster running both |
52 |
>> Gentoo 2005.1 and Mac OS 10.4 (mac os was ~ %5 faster on average). I |
53 |
>> ran IOZone on our XRaid using HFS+ under Mac OS, and Reiser, xfs, |
54 |
>> jfs, |
55 |
>> and ext3 under Gentoo for PPC64. |
56 |
>> |
57 |
>> I also ran tests using Mr Bayes, ClustalW, MPIBench, and several |
58 |
>> other |
59 |
>> in house tests that I had available but all of these tests where |
60 |
>> either inconclusive or failed to demonstrate any difference between |
61 |
>> the platforms. |
62 |
>> |
63 |
>> I will be doing a bunch of this again here in a bit in order to get |
64 |
>> data on our new OS build we are putting into production =) |
65 |
>> |
66 |
>> |
67 |
>> On Apr 11, 2006, at 6:20 PM, Brandon Edens wrote: |
68 |
>> |
69 |
>>>> Yeah, from the numbers it looks as if it would be dependent on the |
70 |
>>>> purpose of the cluster whether OS X or Gentoo would do better. |
71 |
>>>> On ppc, |
72 |
>>>> Gentoo does poorly on the first two benchmarks and also on context |
73 |
>>>> switching. On x86, the first two are more comparable with RH, |
74 |
>>>> but the |
75 |
>>>> others, Gentoo has a small to large advantage over RH, just as |
76 |
>>>> on ppc. |
77 |
>>> |
78 |
>>> First and second benchmarks are dhrystone and whetstone, synthetic |
79 |
>>> benchmarks. |
80 |
>>> I'd be wary of these types of benchmarks; they've been cheated |
81 |
>>> before. I'd like |
82 |
>>> to see some real-world computation benchmarks. |
83 |
>>> |
84 |
>>> Brandon Edens |
85 |
>>> |
86 |
>> |
87 |
>> --gentoo-cluster@g.o mailing list |
88 |
>> |
89 |
>> |
90 |
> |
91 |
> -- |
92 |
> Justin Bronder |
93 |
> University of Maine, Orono |
94 |
> |
95 |
> Advanced Computing Research Lab |
96 |
> 20 Godfrey Dr |
97 |
> Orono, ME 04473 |
98 |
> www.clusters.umaine.edu |
99 |
> |
100 |
> Mathematics Department |
101 |
> 425 Neville Hall |
102 |
> Orono, ME 04469 |
103 |
> |
104 |
> |
105 |
> -- |
106 |
> gentoo-cluster@g.o mailing list |
107 |
> |
108 |
|
109 |
-- |
110 |
gentoo-cluster@g.o mailing list |