| 1 |
Hi all, |
| 2 |
|
| 3 |
I was wondering if there are any sane ways to optimize the performance |
| 4 |
of a Gentoo system. |
| 5 |
Overoptimization (the well known "-O9 -fomgomg" CFLAGS etc.) tends to |
| 6 |
make things unstable, which is of course not what we want. The "easy" |
| 7 |
way out would be buying faster hardware, but that is usually not an |
| 8 |
option ;-) |
| 9 |
|
| 10 |
So ... what can be done to get the stable maximum out of your hardware? |
| 11 |
|
| 12 |
In my experience (x86 centric - do other arches have different |
| 13 |
"problems"?) the following is stable, but not necessarily the optimum: |
| 14 |
- don't overtweak CFLAGS. "-O2 -march=$your_cpu_family" seems to be on |
| 15 |
average the best, -O3 is often slower and can cause bugs |
| 16 |
- don't do anything with ASFLAGS, LDFLAGS. This causes weird random |
| 17 |
breakage (e.g. LDFLAGS="-O1" causes prelink to fail with "absurd" |
| 18 |
errors) and doesn't give a noticeable performance boost |
| 19 |
- check that all IDE disks use DMA mode, otherwise they are limited to |
| 20 |
~16M/s with a huge CPU usage penalty. Sometimes (application-specific) |
| 21 |
increasing the readahead with hdparm gives a huge throughput boost. |
| 22 |
- kernel tweaks like preempt may increase the responsiveness of the |
| 23 |
system, but often reduce throughput and may have unexpected sideeffects |
| 24 |
like random audio stutter as well as random kernel crashes ;-) |
| 25 |
- kernel tweaks like setting swappiness or using a different I/O |
| 26 |
scheduler (CFQ, deadline) should help, but I'm not aware of any "real" |
| 27 |
benchmarks except microbenchmarks (can create 1M files 10% faster!!!!! - |
| 28 |
yes, but how does it behave with a normal workload?) |
| 29 |
- using a "smarter" filesystem can dramatically improve performance at |
| 30 |
the potential cost of reliability. As data on FS reliability is hard to |
| 31 |
find from unbiased sources this becomes a religious issue ... migrating |
| 32 |
from ext3 to reiserfs makes "emerge sync" extremely much faster, but is |
| 33 |
reiserfs sustainable? |
| 34 |
|
| 35 |
Are there any application-specific tweaks (e.g. "use the prefork MPM |
| 36 |
with apache2")? What is known to break things, what has usually |
| 37 |
beneficial behaviour? Are there any useful benchmarks that show the |
| 38 |
performance difference between different settings? |
| 39 |
|
| 40 |
Thanks for your input, |
| 41 |
|
| 42 |
Patrick |
| 43 |
-- |
| 44 |
Stand still, and let the rest of the universe move |
Archives