1 |
On Sun, Aug 22, 2010 at 12:01 PM, Alan Warren <bluemoonshine@×××××.com> wrote: |
2 |
> Hello, |
3 |
> |
4 |
> I am having some system performance issues with this kernel release. I have |
5 |
> a SMP machine (dual xeon nehalem 8 core / 16 threads) with 24gb non-ecc |
6 |
> memory. |
7 |
> |
8 |
> On occasion (seems random so far) my system feels like a Pentium II trying |
9 |
> to cope with Vista. For example, I was in the middle of tar'ing a semi-large |
10 |
> file and noticed all of my apps came to a crawl. Scrolling in firefox, |
11 |
> typing in the terminal, or trying to navigate in my file manager resulted in |
12 |
> breif "pauses" that came in waves. On one occasion my system froze |
13 |
> completely and I had to manually reset the machine. (that was with |
14 |
> 2.6.35-r1) |
15 |
> |
16 |
> I didn't activate anything "new" in this kernel release that I don't |
17 |
> normally activate. ie, no cpuidle driver |
18 |
> |
19 |
> Is there a proper venu for debugging such matters, or should I just wait for |
20 |
> this kernel to go prime-time? |
21 |
> |
22 |
> Thanks for your time, |
23 |
> Alan |
24 |
> |
25 |
|
26 |
Hi Alan, |
27 |
Sorry for the problems. I've seen them also in the recent past. In |
28 |
my case it was on new hardware so I couldn't say it was due to a |
29 |
specific kernel release. |
30 |
|
31 |
1) What happens when you watch top while doing the tar? Do you by any |
32 |
chance see large wait times in top? (Hit '1' to watch all CPUs) If so |
33 |
the problem could well be how the kernel is dealing with writing data |
34 |
back to the hard drive. I had this problem with the WD Green drives. |
35 |
When I changed to WD RAID Edition drives (1/2 the storage for 30% more |
36 |
money) the problems disappeared. |
37 |
|
38 |
2) If it's not the drive issue then there is a kernel option called (I |
39 |
think) RCU_CPU_STALL_DETECTION (or something like that. If you turn it |
40 |
on I may generate a trace of what's keeping a core busy to long. |
41 |
Mileage will vary. |
42 |
|
43 |
Good luck, |
44 |
Mark |