Messages in this thread | | | Date | Fri, 07 Nov 2008 09:29:56 -0500 | From | "Alan D. Brunelle" <> | Subject | Re: [PATCH][v3] blktrace: conversion to tracepoints |
| |
> > I'll apply this for 2.6.29. I'm assuming you have tested this as well? >
FYI: I tested this on a 16-way AMD x86_64 box (128GB RAM, a couple of Smart Array (CCISS) P800's). Running Jens' for-2.6.29 git tree versus a recent Linus 2.6.28-rc3 tree. I was doing kernel makes (10 passes for both kernels), and found that things worked just fine. All values below are in seconds:
2.6.28-rc3:
Part Min Avg Max Dev -------- ------- ------- ------- ------- untar 17.996 22.565 23.522 1.690 mk-cfg 3.279 3.454 3.529 0.070 mk-kern 140.837 141.557 142.549 0.626 mk-clean 10.460 11.604 12.710 0.758 -------- ------- ------- ------- ------- Combd 172.941 179.180 181.600 2.470 ======== ======= ======= ======= ======= %sys 7.67% 7.78% 8.03% 0.114
Jens' origin/for-2.6.29:
Part Min Avg Max Dev -------- ------- ------- ------- ------- untar 18.138 22.576 24.279 1.693 mk-cfg 3.270 3.483 3.751 0.118 mk-kern 141.290 142.115 142.666 0.439 mk-clean 8.263 9.609 10.211 0.624 -------- ------- ------- ------- ------- Combd 173.736 177.783 179.654 1.786 ======== ======= ======= ======= ======= %sys 7.77% 7.85% 8.01% 0.074
Not sure why the 'make clean' portion of the runs went noticeably better with the newer code, but in any event, I don't see much difference elsewhere, and for sure the blktrace portion of things worked correctly. [blktrace + blkparse + blkrawverify + btt all worked just fine out of the box.]
Regards, Alan
| |