On Jun 12, 2009, at 3:52 PM, Styner, Douglas W wrote:
Anirban Chakraborty [mailto:anirban.chakraborty@xxxxxxxxxx] writes:
This is the updated patch. Please apply.
Performance results from this patch compared to 2.6.30-rc6_scsi-misc.
Linux OLTP Performance summary
Kernel# Speedup(x) Intr/s CtxSw/s us% sys% idle%
iowait%
scsi-misc 1.000 29481 43570 74 25
0 0
scsi-misc_qla-irqsave 1.004 29471 43290 75 25
0 0
Server configurations:
Intel Xeon Quad-core 2.0GHz 2 cpus/8 cores/8 threads
64GB memory, 3 qle2462 FC HBA, 450 spindles (30 logical units)
======oprofile CPU_CLK_UNHALTED for top 30 functions
Cycles% 2.6.30-rc6_scsi-misc Cycles% 2.6.30-rc6_scsi-misc_qla-
irqsave
67.3266 <database> 66.6218 <database>
1.0062 qla24xx_start_scsi 1.0247 qla24xx_start_scsi
0.9246 qla24xx_intr_handler 0.9731 qla24xx_intr_handler
0.8158 __schedule 0.8588 kmem_cache_alloc
0.7469 kmem_cache_alloc 0.8515 __schedule
0.4188 __blockdev_direct_IO 0.4663 __sigsetjmp
0.4097 __sigsetjmp 0.4331 __blockdev_direct_IO
0.3989 scsi_request_fn 0.3852 scsi_request_fn
0.3916 __switch_to 0.3852 task_rq_lock
0.3717 __list_add 0.3612 rb_get_reader_page
0.3499 task_rq_lock 0.3336 __switch_to
0.3408 try_to_wake_up 0.3336 ring_buffer_consume
0.3281 aio_complete 0.3299 copy_user_generic_string
0.3281 rb_get_reader_page 0.3262 __list_add
0.3227 ring_buffer_consume 0.3262 lock_timer_base
0.2901 copy_user_generic_string 0.3225 try_to_wake_up
0.2883 <bash> 0.3059 aio_complete
0.2611 blk_queue_end_tag 0.2783 mod_timer
0.2611 memset_c 0.2488 kmem_cache_free
0.2448 kmem_cache_free 0.2451 blk_queue_end_tag
0.2357 qla2x00_process_completed_re0.2451 tcp_sendmsg
0.2321 lock_timer_base 0.2396 <bash>
0.2321 mod_timer 0.2359 kref_get
0.2230 kfree 0.2359 memset_c
0.2230 tcp_sendmsg 0.2304 qla2x00_process_completed_re
0.2176 generic_make_request 0.2285 memmove
0.2085 scsi_dispatch_cmd 0.2285 mempool_free
0.2085 kref_get 0.2248 generic_make_request
0.2085 sched_clock_cpu 0.2156 sched_clock_cpu
0.2067 scsi_device_unbusy 0.2119 e1000_xmit_frame
So, the data that you posted validated earlier didn't match with this
run.
I did a similar testing several times before I posted the patch. Maybe
I should share that data here.
Sever: Intel Xeon X7350, 4 core, 16GB memory, 2 dual port qla2432 and
1 single port qla2432 (total 5 controllers).
Target: EMC Clarion (100+ luns per target, total no. of luns 512).
IO pumping tool: Orion with cold cache setting. Two Orion procs
running each pumping to 256 devices.
Profiling tool: vtune.
And my results are as follows:
2.6.30-rc6 (without the patch)
function CPU_CLK_UNHALTED.CORE_P
qla24xx_intr_handler 1.18
qla2x00_process_completed_requ 0.18
qla24xx_start_scsi 0.15
qla24xx_process_response_queue 0.12
qla2xxx_queuecommand 0.06
qla2x00_sp_compl 0.06
qla2x00_status_entry 0.02
qla2x00_async_event 0.01
and 2.6.30-rc6 (with the irqsave patch)
function CPU_CLK_UNHALTED.CORE_P
qla24xx_start_scsi 0.11
qla2xxx_queuecommand 0.06
qla24xx_intr_handler 0.01
qla2x00_timer 0.00
The difference in qla24xx_intr_handler is significant in my setup.
Thanks,
Anirban
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
