I sent the attached posting to the kernel list a couple of weeks ago and
got no response, not surprising considering how vague I was being.
I had some spare time on Friday so I thought I'd just confirm to myself
that I am seeing this. I built each kernel between 123 to 129, rebooted
and did the compile of my app. There was no variation in compile times, they
all took around 40 seconds. I assumed I'd just been imagining things and
went home for the weekend. When I arrived at work yesterday and did a
rebuild it took 30 seconds !
What I think is happening is the write buffers are small after reboot so
the extra 10 seconds is spent waiting for disk writes to finish. To test
this I ran a script last night that does a rebuild every half hour, it
also recorded the contents of /proc/meminfo.
The full output of this program is on http://194.164.88.34/bench.out
to summarise, after reboot:
27.51user 3.99system 0:43.34elapsed 72%CPU (0avgtext+0avgdata0maxresident)k
0inputs+0outputs (47744major+30281minor)pagefaults 0swaps
total: used: free: shared: buffers: cached:
Mem: 81129472 24109056 57020416 35848192 7147520 8732672
Swap: 131567616 0 131567616
MemTotal: 79228 kB
MemFree: 55684 kB
MemShared: 35008 kB
Buffers: 6980 kB
Cached: 8528 kB
SwapTotal: 128484 kB
SwapFree: 128484 kB
after 10 hours:
27.02user 3.92system 0:41.34elapsed 74%CPU (0avgtext+0avgdata0maxresident)k
0inputs+0outputs (47570major+30281minor)pagefaults 0swaps
total: used: free: shared: buffers: cached:
Mem: 81129472 71888896 9240576 35897344 50335744 8912896
Swap: 131567616 0 131567616
MemTotal: 79228 kB
MemFree: 9024 kB
MemShared: 35056 kB
Buffers: 49156 kB
Cached: 8704 kB
SwapTotal: 128484 kB
SwapFree: 128484 kB
after 10.5 hours:
27.35user 3.40system 0:32.33elapsed 95%CPU (0avgtext+0avgdata0maxresident)k
0inputs+0outputs (46532major+30281minor)pagefaults 0swaps
total: used: free: shared: buffers: cached:
Mem: 81129472 75911168 5218304 35782656 50335744 10694656
Swap: 131567616 8192 131559424
MemTotal: 79228 kB
MemFree: 5096 kB
MemShared: 34944 kB
Buffers: 49156 kB
Cached: 10444 kB
SwapTotal: 128484 kB
SwapFree: 128476 kB
after 11 hours:
27.32user 3.31system 0:31.15elapsed 98%CPU (0avgtext+0avgdata0maxresident)k
0inputs+0outputs (46505major+30281minor)pagefaults 0swaps
total: used: free: shared: buffers: cached:
Mem: 81129472 75952128 5177344 35782656 50335744 10702848
Swap: 131567616 8192 131559424
MemTotal: 79228 kB
MemFree: 5056 kB
MemShared: 34944 kB
Buffers: 49156 kB
Cached: 10452 kB
SwapTotal: 128484 kB
SwapFree: 128476 kB
Does anyone have any suggestions as to why this is happening, or where in
the code I should start my search. Is it the buffers ? Is this a novel
way of making people run Linux as long as possible ? 'Linux, it gets
faster the more you use it'.
Please mail me if you want me to try other experiments.
- Joe Thornber
> I've been using the development kernels from about 105 onwards and am
> puzzled by some performance variations.
> When I've built a new kernel I normally try it out by rebuilding the
> project I'm working on by doing :
> make really_clean; make; make really_clean; time make
> Some of the development kernels take ~30 seconds elapsed time to compile
> my program while others take ~40 seconds. I haven't got exact figures
> except I know that yesterdays 2.1.127 took 31 seconds 94% cpu, and
> todays 2.1.128 took 41 seconds 74% cpu. This was done on the same
> system, just after a reboot. Why the difference in in cpu% ?
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/