Re: 2.6.8-rc2-mm2

[Hideo AOKI]

Hi, 

Rik van Riel wrote:

> I would really appreciate any testing results on this, both good
> and bad.  I want to get this thing tuned and into a generally good
> shape for use by everybody upstream.
>
> I'm especially interested in how this affects compute servers,
> desktops and heavily overloaded network servers (the "spamassassin
> slowed my system to a crawl" symptom would be one to test ;)).
>
> I suspect the patch may need some tweaking to help interactivity
> in some cases, but maybe it'll already work magically by itself...

To evaluate your patch, I tested -mm kernels for thrashing generated
by our benchmark suit.

In 2.6.8-rc2-mm2, performance of application increased, and overhead
of kernel decreased.


** Overview of workload

Benchmark program creates disk I/O processes and shared memory access
processes. Generating disk I/O and shared memory access at the same
time.



** Detailed environment of performance evaluation

- Hardware
  CPU: Xeon 1.6GHz * 4
  Memory: 2GB
  HDD: IDE ATA100 

- The benchmark suit 
  The benchmark suit is the wblg-disk, which I have released on
  SourceForge.

  <http://sourceforge.net/project/showfiles.php?group_id=110454&package_id=119281> 
  # I used alpha version (1.0.3-alpha) in this test.
  # I will release 1.0.3 soon.

- configuration of benchmark
  1. disk IO workload
    + number of process: 4 processes
    + read/write ratio: read 0% / write 100%
    + file IO size: 1KB - 256KB (random) 

  2. memory access workload
    + number of process: 32 processes
    + Shard memory regions: 512MB * 4 
    + Each process accesses one shared memory region.
      (8 processes share one shared memory region.)
    + Each process repeats 4 byte memory access each 4KB.

  Measurement time: 1 hour
  Measurement items: write throughput

- Other configurations
  Using Oprofile 
  Issue vmstat command each 1 minute



** Results of performance evaluation

* result of benchmark
                       write throughput [MB/s]
       2.6.8-rc2-mm1:  2.39
       2.6.8-rc2-mm2:  2.68


* result of vmstat
                       swap in(kB/s)   swap out(kB/s)
       2.6.8-rc2-mm1:  0 - 641         0 - 583         
       2.6.8-rc2-mm2:  0 - 727         0 - 684         


* results of Oprofile

 - System profiling (top 2)

   Overhead of kernel decreased by 3.9%.

   2.6.8-rc2-mm1:
       No.     samples         %               
       1.      170981153       77.6102 wblg-disk_client
       2.       46808551       21.2469 vmlinux-2.6.8-rc2-mm1

   2.6.8-rc2-mm2:
       No.     samples         %               
       1.      179798682       81.6102 wblg-disk_client
       2.       38141112       17.3122 vmlinux-2.6.8-rc2-mm2


 - Kernel profiling (top 10) 

   Ratio of page_referenced_one() and kmap_atomic() decreased.

   2.6.8-rc2-mm1:
       No.     samples         %       symbol name
       1       11398599        24.3515 page_referenced_one
       2        5652276        12.0753 kmap_atomic
       3        2257662         4.8232 sysenter_past_esp
       4        2098275         4.4827 flush_tlb_others
       5        1467288         3.1347 __copy_to_user_ll
       6        1167740         2.4947 page_address
       7        1145108         2.4464 page_fault
       8         946909         2.0229 flush_tlb_page
       9         889151         1.8995 do_gettimeofday
       10        828174         1.7693 refill_inactive_zone

   2.6.8-rc2-mm2:
       No.     samples         %       symbol name
       1        6057840        15.8827 page_referenced_one
       2        3707571         9.7207 kmap_atomic
       3        2531840         6.6381 sysenter_past_esp
       4        1707017         4.4755 __copy_to_user_ll
       5        1328610         3.4834 flush_tlb_others
       6        1233726         3.2346 page_fault
       7        1140917         2.9913 do_gettimeofday
       8         970534         2.5446 get_offset_tsc
       9         759579         1.9915 mark_offset_tsc
       10        650208         1.7047 page_address


Best regards,
Hideo AOKI

Systems Development Laboratory, Hitachi, Ltd.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/