Mainline kernel OLTP performance update

Mainline kernel OLTP performance update

[Styner, Douglas W]

Summary: Measured the mainline kernel from kernel.org (2.6.30-rc2). 

The regression for 2.6.30-rc2 against the baseline, 2.6.24.2 is 1.95%.  Note the dip in cycles for database compared to us% in summary.  

Linux OLTP Performance summary
Kernel#            Speedup(x)   Intr/s  CtxSw/s us%     sys%    idle%   iowait%
2.6.24.2                1.000   22106   43709   75      24      0       0
2.6.30-rc2              0.981   30755   43072   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 0.9.3 CPU_CLK_UNHALTED for top 30 functions
Cycles% 2.6.24.2                   Cycles% 2.6.30-rc2
74.8578 <database>                   67.6966 <database>
1.0500 qla24xx_start_scsi          1.1724 qla24xx_start_scsi
0.8089 schedule                    1.0578 qla24xx_intr_handler
0.5864 kmem_cache_alloc            0.8259 __schedule
0.4989 __blockdev_direct_IO        0.7451 kmem_cache_alloc
0.4357 __sigsetjmp                 0.4872 __blockdev_direct_IO
0.4152 copy_user_generic_string    0.4390 task_rq_lock
0.3953 qla24xx_intr_handler        0.4338 __sigsetjmp
0.3850 memcpy                      0.4195 __switch_to
0.3596 scsi_request_fn             0.3713 copy_user_generic_string
0.3188 __switch_to                 0.3608 __list_add
0.2889 lock_timer_base             0.3595 rb_get_reader_page
0.2750 memmove                     0.3309 ring_buffer_consume
0.2519 task_rq_lock                0.3152 scsi_request_fn
0.2474 aio_complete                0.3048 try_to_wake_up
0.2460 scsi_alloc_sgtable          0.2983 tcp_sendmsg
0.2445 generic_make_request        0.2931 lock_timer_base
0.2263 qla2x00_process_completed_re0.2840 aio_complete
0.2118 blk_queue_end_tag           0.2697 memset_c
0.2085 dio_bio_complete            0.2527 mod_timer
0.2021 e1000_xmit_frame            0.2462 qla2x00_process_completed_re
0.2006 __end_that_request_first    0.2449 memmove
0.1954 generic_file_aio_read       0.2358 blk_queue_end_tag
0.1949 kfree                       0.2241 generic_make_request
0.1915 tcp_sendmsg                 0.2215 scsi_device_unbusy
0.1901 try_to_wake_up              0.2162 mempool_free
0.1895 kref_get                    0.2097 e1000_xmit_frame
0.1864 __mod_timer                 0.2097 kmem_cache_free
0.1863 thread_return               0.2058 kfree
0.1854 math_state_restore          0.1993 sched_clock_cpu
 
Thanks,
Doug

Re: Mainline kernel OLTP performance update

[Andi Kleen]

"Styner, Douglas W" <douglas.w.styner@intel.com> writes:

>
> ======oprofile 0.9.3 CPU_CLK_UNHALTED for top 30 functions
> Cycles% 2.6.24.2                   Cycles% 2.6.30-rc2
> 74.8578 <database>                   67.6966 <database>

The dip in database cycles is indeed worrying.

> 1.0500 qla24xx_start_scsi          1.1724 qla24xx_start_scsi
> 0.8089 schedule                    1.0578 qla24xx_intr_handler
> 0.5864 kmem_cache_alloc            0.8259 __schedule
> 0.4989 __blockdev_direct_IO        0.7451 kmem_cache_alloc
> 0.4357 __sigsetjmp                 0.4872 __blockdev_direct_IO
> 0.4152 copy_user_generic_string    0.4390 task_rq_lock
> 0.3953 qla24xx_intr_handler        0.4338 __sigsetjmp

And also why the qla24xx_intr_handler became ~2.5x as expensive.
Cc linux-scsi and qla24xx maintainers.

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only.

Discrepancies between Oprofile and vmstat

[Styner, Douglas W]

We are using an OLTP-type workload on a fully saturated system and have begun seeing a discrepancy between the output of oprofile and vmstat on upstream kernels that was first seen with the 2.6.27 kernel.

Workload:
OLTP database workload using QLogic FC cards with MSI-X enabled doing ~50,000 IOPS

Oprofile (v0.9.3) collection methodology:
opcontrol --init
opcontrol --vmlinux=/usr/src/linux-`uname -r`/vmlinux
opcontrol -e=CPU_CLK_UNHALTED:80000 -e=LLC_MISSES:6000
opcontrol --start-daemon
sleep 5
opcontrol --start
sleep 60
opcontrol --stop
opcontrol --save=$1
opcontrol --deinit

Function classification:
Modules vmlinux and *.ko = kernel, all others = user.

Results (2.6.30-rc3 kernel):
vmstat reports ~75% user, 25% kernel
oprofile reports 70.3808% user, 29.6142% kernel 

Using another profiling tool to confirm, we see 74.784% user, 25.174% kernel.

Doug Styner
douglas.w.styner@intel.com

Re: Discrepancies between Oprofile and vmstat

[Andi Kleen]

On Mon, Apr 27, 2009 at 11:38:08AM -0700, Styner, Douglas W wrote:
> We are using an OLTP-type workload on a fully saturated system and have begun seeing a discrepancy between the output of oprofile and vmstat on upstream kernels that was first seen with the 2.6.27 kernel.

Rafael,

something to track for the regression list.

-Andi

RE: Discrepancies between Oprofile and vmstat II

[Styner, Douglas W]

I believe so, but will confirm.

> opcontrol -e=CPU_CLK_UNHALTED:80000 -e=LLC_MISSES:6000
> 
> Using another profiling tool to confirm, we see 74.784% user, 25.174% kernel.

Just verifying -- you also see it when you use a shorter period than 80000 right?

Re: Discrepancies between Oprofile and vmstat II

[Andi Kleen]

> opcontrol -e=CPU_CLK_UNHALTED:80000 -e=LLC_MISSES:6000
> 
> Using another profiling tool to confirm, we see 74.784% user, 25.174% kernel.

Just verifying -- you also see it when you use a shorter period than 80000 right?

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only.

RE: Discrepancies between Oprofile and vmstat II

[Styner, Douglas W]

> I believe so, but will confirm.
>
> > opcontrol -e=CPU_CLK_UNHALTED:80000 -e=LLC_MISSES:6000
> > 
> > Using another profiling tool to confirm, we see 74.784% user, 25.174% > > kernel.
>
> Just verifying -- you also see it when you use a shorter period than 80000 > right?

Confirmed.  Changes in profile appear to be due to increased sampling by oprofile with lower value.

Linux OLTP Performance summary
Kernel#            Speedup(x)   Intr/s  CtxSw/s us%     sys%    idle%   iowait%
2.6.30-rc3              1.000   30593   43976   74      25      0       1
2.6.30-rc3              1.001   30534   43210   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
-e=CPU_CLK_UNHALTED:80000          -e=CPU_CLK_UNHALTED:20000
Cycles% 2.6.30-rc3                 Cycles% 2.6.30-rc3
68.5544 <database>                 65.6876 <database>
1.1859 qla24xx_start_scsi          1.1033 kstat_irqs_cpu
0.9436 qla24xx_intr_handler        1.0575 rb_get_reader_page
0.8307 __schedule                  1.0034 qla24xx_start_scsi
0.7194 kmem_cache_alloc            0.9305 qla24xx_intr_handler
0.5026 __blockdev_direct_IO        0.8410 ring_buffer_consume
0.4319 __sigsetjmp                 0.8160 __schedule
0.4244 scsi_request_fn             0.5683 kmem_cache_alloc
0.3853 rb_get_reader_page          0.4517 __sigsetjmp
0.3777 __switch_to                 0.4413 unmap_vmas
0.3552 __list_add                  0.3809 __blockdev_direct_IO
0.3552 task_rq_lock                0.3726 __switch_to
0.3371 try_to_wake_up              0.3310 __list_add
0.3221 ring_buffer_consume         0.3206 task_rq_lock
0.2844 aio_complete                0.3123 scsi_request_fn
0.2588 memmove                     0.2977 aio_complete
0.2588 mod_timer                   0.2914 try_to_wake_up
0.2558 generic_make_request        0.2644 page_fault
0.2543 tcp_sendmsg                 0.2436 kmem_cache_free
0.2528 copy_user_generic_string    0.2415 scsi_device_unbusy
0.2468 lock_timer_base             0.2352 copy_user_generic_string
0.2468 memset_c                    0.2311 memmove
0.2288 blk_queue_end_tag           0.2227 lock_timer_base
0.2257 qla2x00_process_completed_re0.2123 generic_make_request
0.2212 kref_get                    0.2103 kfree
0.2182 mempool_free                0.2040 find_vma
0.2137 sd_prep_fn                  0.2019 sd_prep_fn
0.2122 e1000_xmit_frame            0.2019 blk_queue_end_tag
0.2062 dequeue_rt_stack            0.2019 tcp_sendmsg
0.2047 scsi_device_unbusy          0.1998 qla2x00_process_completed_re

Re: Discrepancies between Oprofile and vmstat II

[Andi Kleen]

On Mon, Apr 27, 2009 at 03:15:27PM -0700, Styner, Douglas W wrote:
> 
> > I believe so, but will confirm.
> >
> > > opcontrol -e=CPU_CLK_UNHALTED:80000 -e=LLC_MISSES:6000
> > > 
> > > Using another profiling tool to confirm, we see 74.784% user, 25.174% > > kernel.
> >
> > Just verifying -- you also see it when you use a shorter period than 80000 > right?
> 
> Confirmed.  Changes in profile appear to be due to increased sampling by oprofile with lower value.

You mean "with higher value", right ?

It's still strange that it suddenly changed in 2.6.27 though. There shouldn't have been
a difference there. So I still think it's a regression.

-Andi

RE: Discrepancies between Oprofile and vmstat II

[Styner, Douglas W]

>
>You mean "with higher value", right ?

Right.  Higher number of samples; lower count value.  I'm going to try it the other way as well (count value = 200000)

>
>It's still strange that it suddenly changed in 2.6.27 though. There
>shouldn't have been
>a difference there. So I still think it's a regression.
>

We only first saw it in 2.6.27.  It is possible that it goes back to 2.6.26 or 2.6.25.  We have not confirmed those kernels.

Re: Discrepancies between Oprofile and vmstat II

[Eric Dumazet]

Andi Kleen a écrit :
> On Mon, Apr 27, 2009 at 03:15:27PM -0700, Styner, Douglas W wrote:
>>> I believe so, but will confirm.
>>>
>>>> opcontrol -e=CPU_CLK_UNHALTED:80000 -e=LLC_MISSES:6000
>>>>
>>>> Using another profiling tool to confirm, we see 74.784% user, 25.174% > > kernel.
>>> Just verifying -- you also see it when you use a shorter period than 80000 > right?
>> Confirmed.  Changes in profile appear to be due to increased sampling by oprofile with lower value.
> 
> You mean "with higher value", right ?
> 
> It's still strange that it suddenly changed in 2.6.27 though. There shouldn't have been
> a difference there. So I still think it's a regression.
> 

An oprofile regression ? Oh no !

;) Just kidding

Introduction of ring buffers was probably the start of oprofile using more cpu cycles,
and I remember seeing rb_get_reader_page() and ring_buffer_consume()
climbing in top 10 functions on tbench oprofile session.

CPU: Core 2, speed 3000.46 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a unit mask of 0x00 (Unhalted core cycles) count 10000
samples  cum. samples  %        cum. %     symbol name
731641   731641        10.0572  10.0572    copy_from_user
328634   1060275        4.5174  14.5746    copy_to_user
269686   1329961        3.7071  18.2817    ipt_do_table
254019   1583980        3.4918  21.7735    __schedule
205152   1789132        2.8200  24.5935    tcp_ack
202930   1992062        2.7895  27.3830    rb_get_reader_page   rank 6
170751   2162813        2.3472  29.7301    tcp_sendmsg
152271   2315084        2.0931  31.8233    ring_buffer_consume  rank 8
147362   2462446        2.0256  33.8489    tcp_recvmsg
138150   2600596        1.8990  35.7479    tcp_transmit_skb
134191   2734787        1.8446  37.5925    ip_queue_xmit
125952   2860739        1.7313  39.3239    sysenter_past_esp
112158   2972897        1.5417  40.8656    tcp_rcv_established
106236   3079133        1.4603  42.3259    __switch_to
105435   3184568        1.4493  43.7752    tcp_v4_rcv
97660    3282228        1.3424  45.1177    netif_receive_skb
89327    3371555        1.2279  46.3456    dst_release
78746    3450301        1.0824  47.4280    dev_queue_xmit
74776    3525077        1.0279  48.4559    release_sock
72979    3598056        1.0032  49.4591    lock_sock_nested
72031    3670087        0.9901  50.4492    tcp_event_data_recv
71130    3741217        0.9778  51.4270    tcp_write_xmit
69475    3810692        0.9550  52.3820    __inet_lookup_established
66537    3877229        0.9146  53.2966    nf_iterate
60504    3937733        0.8317  54.1283    ip_rcv
55767    3993500        0.7666  54.8949    sys_socketcall
53273    4046773        0.7323  55.6272    update_curr
49315    4096088        0.6779  56.3051    ip_finish_output
48958    4145046        0.6730  56.9780    __alloc_skb
48440    4193486        0.6659  57.6439    skb_release_data
46166    4239652        0.6346  58.2785    page_address
46147    4285799        0.6343  58.9128    local_bh_enable
43867    4329666        0.6030  59.5158    netif_rx
43671    4373337        0.6003  60.1161    put_page
43554    4416891        0.5987  60.7148    __update_sched_clock
42899    4459790        0.5897  61.3045    get_page_from_freelist
42747    4502537        0.5876  61.8921    sched_clock_cpu


Also other rb functions use some cycles...

c0261580 <rb_advance_reader>: /* rb_advance_reader total:  95595  0.4967 */
c0261cf0 <rb_buffer_peek>: /* rb_buffer_peek total:  75520  0.3924 */
c0260d80 <rb_event_length>: /* rb_event_length total:  70396  0.3658 */
c0260dd0 <ring_buffer_event_length>: /* ring_buffer_event_length total:  31028  0.1612 */
c02613b0 <ring_buffer_event_data>: /* ring_buffer_event_data total:  20663  0.1074 */

Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a unit mask of 0x00 (Unhalted core cycles) count 10000
CPU_CLK_UNHALT...|
  samples|      %|
------------------
 12062103 80.0850 vmlinux
  1319946  8.7636 libc-2.3.4.so
  1089407  7.2330 tbench
        CPU_CLK_UNHALT...|
          samples|      %|
        ------------------
          1089406 99.9999 tbench
                1 9.2e-05 [heap] (tgid:16048 range:0x804e000-0x8080000)
   403877  2.6815 oprofile
   107343  0.7127 tbench_srv
    55018  0.3653 oprofiled

c02613e0 <rb_get_reader_page>: /* rb_get_reader_page total: 534995  2.7800 */
 13316  0.0692 :c02613e0:       push   %ebp
     5 2.6e-05 :c02613e1:       mov    %esp,%ebp
               :c02613e3:       push   %edi
  6864  0.0357 :c02613e4:       push   %esi
    11 5.7e-05 :c02613e5:       push   %ebx
    14 7.3e-05 :c02613e6:       sub    $0x1c,%esp
  6738  0.0350 :c02613e9:       pushf
 27140  0.1410 :c02613ea:       popl   0x10(%esp)
 13627  0.0708 :c02613ee:       cli
 27016  0.1404 :c02613ef:       lea    0xc(%eax),%edx
    17 8.8e-05 :c02613f2:       mov    %edx,0x14(%esp)
    16 8.3e-05 :c02613f6:       mov    $0x100,%edx
  6765  0.0352 :c02613fb:       mov    0x14(%esp),%ecx
    20 1.0e-04 :c02613ff:       lock xadd %dx,(%ecx)
142726  0.7417 :c0261404:       cmp    %dh,%dl
    11 5.7e-05 :c0261406:       je     c026140e <rb_get_reader_page+0x2e>
               :c0261408:       pause
               :c026140a:       mov    (%ecx),%dl
               :c026140c:       jmp    c0261404 <rb_get_reader_page+0x24>
  6724  0.0349 :c026140e:       mov    0x24(%eax),%ebx
 67811  0.3524 :c0261411:       mov    0x4(%ebx),%edx
 27227  0.1415 :c0261414:       mov    0x10(%ebx),%ecx
  6908  0.0359 :c0261417:       mov    0x8(%ecx),%esi
 21286  0.1106 :c026141a:       cmp    %esi,%edx
     4 2.1e-05 :c026141c:       jb     c0261528 <rb_get_reader_page+0x148>
     8 4.2e-05 :c0261422:       mov    0x8(%ecx),%ecx
   197  0.0010 :c0261425:       cmp    %ecx,%edx
    13 6.8e-05 :c0261427:       ja     c02614fa <rb_get_reader_page+0x11a>
    12 6.2e-05 :c026142d:       cmp    %ebx,0x20(%eax)
    12 6.2e-05 :c0261430:       je     c0261540 <rb_get_reader_page+0x160>
    13 6.8e-05 :c0261436:       lea    0x10(%eax),%esi
    95 4.9e-04 :c0261439:       mov    $0x1,%edx
               :c026143e:       mov    %esi,0x18(%esp)
    45 2.3e-04 :c0261442:       jmp    c0261494 <rb_get_reader_page+0xb4>
               :c0261444:       lea    0x0(%esi,%eiz,1),%esi
    24 1.2e-04 :c0261448:       sub    $0x8,%ebx
    43 2.2e-04 :c026144b:       mov    %ebx,0x18(%eax)
    36 1.9e-04 :c026144e:       mov    %ecx,0x24(%eax)
    80 4.2e-04 :c0261451:       inc    %edx
               :c0261452:       mov    0x10(%ecx),%ebx
               :c0261455:       cmp    $0x4,%edx
...
  6952  0.0361 :c0261528:       mov    0x14(%esp),%edx
   203  0.0011 :c026152c:       incb   0xc(%eax)
  6789  0.0353 :c026152f:       pushl  0x10(%esp)
    30 1.6e-04 :c0261533:       popf
108826  0.5655 :c0261534:       add    $0x1c,%esp
   322  0.0017 :c0261537:       mov    %ebx,%eax
               :c0261539:       pop    %ebx
 13921  0.0723 :c026153a:       pop    %esi
  6689  0.0348 :c026153b:       pop    %edi
  6884  0.0358 :c026153c:       leave
  6791  0.0353 :c026153d:       ret
               :c026153e:       xchg   %ax,%ax
     2 1.0e-05 :c0261540:       xor    %ebx,%ebx
     7 3.6e-05 :c0261542:       jmp    c0261528 <rb_get_reader_page+0x148>
               :c0261544:       lea    0x40(%eax),%edx


c0261de0 <ring_buffer_consume>: /* ring_buffer_consume total: 395772  2.0566 */
  6854  0.0356 :c0261de0:       push   %ebp
    26 1.4e-04 :c0261de1:       mov    %esp,%ebp
               :c0261de3:       push   %edi
 12755  0.0663 :c0261de4:       mov    %eax,%edi
               :c0261de6:       push   %esi
   182 9.5e-04 :c0261de7:       add    $0x10,%edi
               :c0261dea:       push   %ebx
  5464  0.0284 :c0261deb:       xor    %esi,%esi
               :c0261ded:       sub    $0x18,%esp
    42 2.2e-04 :c0261df0:       mov    %edx,%ebx
  3494  0.0182 :c0261df2:       mov    %eax,-0x18(%ebp)
   301  0.0016 :c0261df5:       mov    %ecx,-0x24(%ebp)
   610  0.0032 :c0261df8:       lea    0x0(,%edx,4),%eax
               :c0261dff:       mov    %eax,-0x20(%ebp)
  3672  0.0191 :c0261e02:       jmp    c0261e13 <ring_buffer_consume+0x33>
               :c0261e04:       lea    0x0(%esi,%eiz,1),%esi
               :c0261e08:       test   %esi,%esi
               :c0261e0a:       je     c0261e68 <ring_buffer_consume+0x88>
  3367  0.0175 :c0261e0c:       testb  $0x3,(%esi)
    37 1.9e-04 :c0261e0f:       jne    c0261e68 <ring_buffer_consume+0x88>
               :c0261e11:       pause
  3410  0.0177 :c0261e13:       bt     %ebx,(%edi)
  7766  0.0404 :c0261e16:       sbb    %eax,%eax
  4221  0.0219 :c0261e18:       test   %eax,%eax
               :c0261e1a:       je     c0261e08 <ring_buffer_consume+0x28>
   940  0.0049 :c0261e1c:       mov    -0x18(%ebp),%edx
  2967  0.0154 :c0261e1f:       mov    0x28(%edx),%eax
 12178  0.0633 :c0261e22:       mov    -0x20(%ebp),%edx
  2547  0.0132 :c0261e25:       mov    (%eax,%edx,1),%eax
  5736  0.0298 :c0261e28:       mov    %eax,-0x14(%ebp)
  1065  0.0055 :c0261e2b:       add    $0x8,%eax
 27718  0.1440 :c0261e2e:       mov    %eax,-0x10(%ebp)
  4312  0.0224 :c0261e31:       call   c04d23b0 <_spin_lock_irqsave>
183800  0.9551 :c0261e36:       mov    -0x24(%ebp),%ecx
  3613  0.0188 :c0261e39:       mov    %eax,-0x1c(%ebp)
  6656  0.0346 :c0261e3c:       mov    %ebx,%edx
               :c0261e3e:       mov    -0x18(%ebp),%eax
    11 5.7e-05 :c0261e41:       call   c0261cf0 <rb_buffer_peek>
  3506  0.0182 :c0261e46:       mov    %eax,%esi
     2 1.0e-05 :c0261e48:       test   %eax,%eax
     2 1.0e-05 :c0261e4a:       je     c0261e54 <ring_buffer_consume+0x74>
     2 1.0e-05 :c0261e4c:       mov    -0x14(%ebp),%eax
    10 5.2e-05 :c0261e4f:       call   c0261580 <rb_advance_reader>
  3389  0.0176 :c0261e54:       mov    -0x1c(%ebp),%edx
    13 6.8e-05 :c0261e57:       mov    -0x10(%ebp),%eax
    17 8.8e-05 :c0261e5a:       call   c04d2500 <_spin_unlock_irqrestore>
 71541  0.3718 :c0261e5f:       test   %esi,%esi
     4 2.1e-05 :c0261e61:       jne    c0261e0c <ring_buffer_consume+0x2c>
               :c0261e63:       nop
               :c0261e64:       lea    0x0(%esi,%eiz,1),%esi
  3419  0.0178 :c0261e68:       add    $0x18,%esp
    11 5.7e-05 :c0261e6b:       mov    %esi,%eax
     1 5.2e-06 :c0261e6d:       pop    %ebx
  3403  0.0177 :c0261e6e:       pop    %esi
    17 8.8e-05 :c0261e6f:       pop    %edi
  3343  0.0174 :c0261e70:       leave
  3348  0.0174 :c0261e71:       ret

Re: Mainline kernel OLTP performance update

[Chuck Ebbert]

On Mon, 27 Apr 2009 09:02:40 +0200
Andi Kleen <andi@firstfloor.org> wrote:

> "Styner, Douglas W" <douglas.w.styner@intel.com> writes:
> 
> >
> > ======oprofile 0.9.3 CPU_CLK_UNHALTED for top 30 functions
> > Cycles% 2.6.24.2                   Cycles% 2.6.30-rc2
> > 74.8578 <database>                   67.6966 <database>
> 
> The dip in database cycles is indeed worrying.
> 
> > 1.0500 qla24xx_start_scsi          1.1724 qla24xx_start_scsi
> > 0.8089 schedule                    1.0578 qla24xx_intr_handler
> > 0.5864 kmem_cache_alloc            0.8259 __schedule
> > 0.4989 __blockdev_direct_IO        0.7451 kmem_cache_alloc
> > 0.4357 __sigsetjmp                 0.4872 __blockdev_direct_IO
> > 0.4152 copy_user_generic_string    0.4390 task_rq_lock
> > 0.3953 qla24xx_intr_handler        0.4338 __sigsetjmp
> 
> And also why the qla24xx_intr_handler became ~2.5x as expensive.
> Cc linux-scsi and qla24xx maintainers.
> 

They are getting 31000 interrupts/sec vs. 22000/sec on older kernels.

Re: Mainline kernel OLTP performance update

[James Bottomley]

On Tue, 2009-04-28 at 12:57 -0400, Chuck Ebbert wrote:
> On Mon, 27 Apr 2009 09:02:40 +0200
> Andi Kleen <andi@firstfloor.org> wrote:
> 
> > "Styner, Douglas W" <douglas.w.styner@intel.com> writes:
> > 
> > >
> > > ======oprofile 0.9.3 CPU_CLK_UNHALTED for top 30 functions
> > > Cycles% 2.6.24.2                   Cycles% 2.6.30-rc2
> > > 74.8578 <database>                   67.6966 <database>
> > 
> > The dip in database cycles is indeed worrying.
> > 
> > > 1.0500 qla24xx_start_scsi          1.1724 qla24xx_start_scsi
> > > 0.8089 schedule                    1.0578 qla24xx_intr_handler
> > > 0.5864 kmem_cache_alloc            0.8259 __schedule
> > > 0.4989 __blockdev_direct_IO        0.7451 kmem_cache_alloc
> > > 0.4357 __sigsetjmp                 0.4872 __blockdev_direct_IO
> > > 0.4152 copy_user_generic_string    0.4390 task_rq_lock
> > > 0.3953 qla24xx_intr_handler        0.4338 __sigsetjmp
> > 
> > And also why the qla24xx_intr_handler became ~2.5x as expensive.
> > Cc linux-scsi and qla24xx maintainers.
> > 
> 
> They are getting 31000 interrupts/sec vs. 22000/sec on older kernels.

Should be fixed by:

http://marc.info/?l=linux-scsi&m=124093712114937

If someone could verify, I'd be grateful.

Thanks,

James

RE: Mainline kernel OLTP performance update

[Styner, Douglas W]

Working on it as we speak...

>-----Original Message-----
>From: James Bottomley [mailto:James.Bottomley@HansenPartnership.com]
>Sent: Tuesday, April 28, 2009 10:16 AM
>To: Chuck Ebbert
>Cc: Andi Kleen; Styner, Douglas W; linux-kernel@vger.kernel.org; linux-
>driver@qlogic.com; linux-scsi@vger.kernel.org
>Subject: Re: Mainline kernel OLTP performance update
>
>On Tue, 2009-04-28 at 12:57 -0400, Chuck Ebbert wrote:
>> On Mon, 27 Apr 2009 09:02:40 +0200
>> Andi Kleen <andi@firstfloor.org> wrote:
>>
>> > "Styner, Douglas W" <douglas.w.styner@intel.com> writes:
>> >
>> > >
>> > > ======oprofile 0.9.3 CPU_CLK_UNHALTED for top 30 functions
>> > > Cycles% 2.6.24.2                   Cycles% 2.6.30-rc2
>> > > 74.8578 <database>                   67.6966 <database>
>> >
>> > The dip in database cycles is indeed worrying.
>> >
>> > > 1.0500 qla24xx_start_scsi          1.1724 qla24xx_start_scsi
>> > > 0.8089 schedule                    1.0578 qla24xx_intr_handler
>> > > 0.5864 kmem_cache_alloc            0.8259 __schedule
>> > > 0.4989 __blockdev_direct_IO        0.7451 kmem_cache_alloc
>> > > 0.4357 __sigsetjmp                 0.4872 __blockdev_direct_IO
>> > > 0.4152 copy_user_generic_string    0.4390 task_rq_lock
>> > > 0.3953 qla24xx_intr_handler        0.4338 __sigsetjmp
>> >
>> > And also why the qla24xx_intr_handler became ~2.5x as expensive.
>> > Cc linux-scsi and qla24xx maintainers.
>> >
>>
>> They are getting 31000 interrupts/sec vs. 22000/sec on older kernels.
>
>Should be fixed by:
>
>http://marc.info/?l=linux-scsi&m=124093712114937
>
>If someone could verify, I'd be grateful.
>
>Thanks,
>
>James
>