1 |
Hi there, |
2 |
|
3 |
On 4/29/07, Peter A. H. Peterson <pedro@×××××××××××.net> wrote: |
4 |
> Hi Everyone, |
5 |
> |
6 |
> My name is Peter Peterson and I represent a group of a grad students |
7 |
> at UCLA. We're in a computer systems performance analysis course and we |
8 |
> were hoping to do a general performance comparison of gentoo vs. a |
9 |
> popular binary i386-compatible distribution (probably ubuntu) in some |
10 |
> "real-world" server tests to try and meaningfully calculate the |
11 |
> performance gains that local compilation provides. (For example, |
12 |
> apache2 requests processed per second on the same hardware.) |
13 |
> |
14 |
> I've subscribed to this list because we want the gentoo community to |
15 |
> be involved in helping us design the tests so that we can hopefully |
16 |
> all feel good about what and how we are testing the systems. |
17 |
> |
18 |
> We have no particular outcome in mind; our group represents a wide |
19 |
> range of computer users, from Mac, Linux, and Windows enthusiasts, and |
20 |
> we have all used a wide variety of Linux distributions. We have simply |
21 |
> noticed that much of the discussion of gentoo's performance advantage |
22 |
> is anecdotal and we're genuinely hoping to provide some meaningful |
23 |
> experimental data for discussion. Also, if anyone knows of any |
24 |
> available benchmark data or papers on this subject, we'd love to hear |
25 |
> about them. There was apparently a paper on slashdot a couple of years |
26 |
> ago, but the host it was on appears to now be squatted. For that |
27 |
> matter, if this is a well understood or closed issue (for example, if |
28 |
> the statistics that people quote are actually from good experimental |
29 |
> data) please let us know. |
30 |
> |
31 |
> Is anyone here interested in discussing this project? We are |
32 |
> specifically interested in discussing methodology, testing suits, |
33 |
> CFLAGS and other options. Our desire is not to "trick out" gentoo or |
34 |
> ubuntu, but rather quantify the performance benefit that gentoo has |
35 |
> over binary distributions with "normal" compile flags (whatever normal |
36 |
> is). |
37 |
> |
38 |
|
39 |
A good CFLAGS would be something not very agressive, something like: |
40 |
-march=<cpuType> -O3 or -O2 and at most -fomit-frame-pointer. |
41 |
(Scientific workloads can speedup considerably with: -ffast-math) |
42 |
|
43 |
Having experienced and done some benchmarks with gentoo and other |
44 |
distros on servers and on scientific workstations. |
45 |
What I found is that sometimes gentoo lacks critical performance |
46 |
patches in glibc that are applied to mainstream distros (redhat, |
47 |
suse..etc) that provide boosts in memcpy, memset, etc..(I remmember a |
48 |
discussion about that some years ago). |
49 |
What I also found out is that the compiler flags only affect |
50 |
workloads that are very compute intensive. not something that depends |
51 |
almost completely on FSB load or IO load.. like most server |
52 |
workloads... -O3 doesn't do much to a working set full of |
53 |
unpredictable branches (like server workloads usually are) and low IPC |
54 |
rate. |
55 |
|
56 |
I really do believe performance boost from gentoo to be practically |
57 |
negligible. The difference will only be apreciable in very few corner |
58 |
cases. Most distros also optimize critical aplications such has: |
59 |
openssl, mplayer.. reducing the possible corner cases. |
60 |
|
61 |
Anyways, doing a "academic" benchmark would be a good idea. |
62 |
|
63 |
Something like: |
64 |
micro-benchmarks: |
65 |
- stream (mem bandwith benchmark) |
66 |
- ?? |
67 |
|
68 |
macro-benchmarks: |
69 |
- apache2 + gzip + php(make it cpu intensive, not IO intensive) |
70 |
- xmlmark ? |
71 |
- kernelbench |
72 |
- pybench ? |
73 |
- openssl bench |
74 |
|
75 |
|
76 |
about methodology: |
77 |
- same system, same bios version, same disks. |
78 |
- All OSes must be installed in the same disk partitions. |
79 |
- the will be trouble about the kernel config: |
80 |
- for mainstream distros you should use the kernel that is provided. |
81 |
- for gentoo, gentoo-sources configured by someone which is |
82 |
experienced, and informed about configuration impacts on performance |
83 |
(ideally a kernel hacker?). |
84 |
|
85 |
- should use the stable versions in gentoo portage? |
86 |
- or should use the same application versions used on mainstream distro? |
87 |
|
88 |
|
89 |
|
90 |
-- |
91 |
Miguel Sousa Filipe |
92 |
-- |
93 |
gentoo-performance@g.o mailing list |