binary tracing

Gui,Jian guij@cn.ibm.com
Mon Apr 3 13:53:00 GMT 2006


I did testing on ppc64/2.6.9-34.EL with the new benchmark framework.
The result also looks fine.

+--- S = Script, R = Runtime
|+-- * = Relayfs                Threads
|| NAME                             1       2       4       8
S : empty probe                  1198    1270    1325     786
R : empty probe                   981    1070    1112     670
S*: printf 100 chars             3017    3054    2724    1916
R*: printf 100 chars             2957    2913    2535    1476
R*: write 100 chars              3167    3113    2692    1480
R*: printf 5 integers            5486    5402    5160    3838
R*: bin_write 5 integers         1687    1759    1860    1107
R*: printf 10 integers          10048    9623    9057    6908
R*: bin_write 10 integers        2461    2377    2477    1431

Here is the binary trace patch and the test script I used. 
The patch should be used against snapshot-20060401. 
It is modified based on the Tom's code on Bug2307 
(http://sourceware.org/bugzilla/attachment.cgi?id=937&action=view). 

I mainly modified the prototype to:
1) fix the system-crash bug when _stp_*print() are used;
2) fix the system-crash bug when pre-cpu file is larger than 1Mbyte;
3) add support of %ld, %lld.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bin_write2.20060403.patch
Type: application/octet-stream
Size: 23880 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/systemtap/attachments/20060403/ba859409/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bin_bench
Type: application/octet-stream
Size: 2324 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/systemtap/attachments/20060403/ba859409/attachment-0001.obj>


More information about the Systemtap mailing list