Skip to content

Limiting scsi.d output to IOs that take a long time.

March 20, 2009

I have added support for only reporting on scsi packets that take more than a certain amount of time to complete to scsi.d. This is partularly useful when combined with the PERF_REPORT option to see just the IO requests that took over a time threshold and still collect statistics on all the other IO requests.

To use this you have to specify “-D REPORT_OVERTIME=X” where X is the time you wish to report on in nano seconds. Here is an example that will only print the details of IO requests that took more than 250ms:

# scsi.d -D REPORT_OVERTIME=$((250*1000*1000)) -D PERF_REPORT  
Only reporting IOs longer than 250000000ns 
Hit Control C to interrupt 
00005.388800363 glm0:-> 0x2a WRITE(10) address 00:00, lba 0x00270e67, len 0x000008, control 0x00 timeout 60 CDBP 300002df438 1 sched(0) cdb(10) 2a0000270e6700000800 
00005.649494475 glm0:<- 0x2a WRITE(10) address 00:00, lba 0x00270e67, len 0x000008, control 0x00 timeout 60 CDBP 300002df438, reason 0x0 (COMPLETED) pkt_state 0x1f state 0x0 Success Time 260833us 
00005.384612799 glm0:-> 0x0a  WRITE(6) address 00:00, lba 0x048564, len 0x000001, control 0x00 timeout 60 CDBP 30002541ac0 1 sched(0) cdb(6) 0a0485640100 
00005.716416658 glm0:<- 0x0a  WRITE(6) address 00:00, lba 0x048564, len 0x000001, control 0x00 timeout 60 CDBP 30002541ac0, reason 0x0 (COMPLETED) pkt_state 0x1f state 0x0 Success Time 331939us 
00005.385907691 glm0:-> 0x0a  WRITE(6) address 00:00, lba 0x0605b4, len 0x000001, control 0x00 timeout 60 CDBP 300035637a0 1 sched(0) cdb(6) 0a0605b40100 
00005.773925990 glm0:<- 0x0a  WRITE(6) address 00:00, lba 0x0605b4, len 0x000001, control 0x00 timeout 60 CDBP 300035637a0, reason 0x0 (COMPLETED) pkt_state 0x1f state 0x0 Success Time 388153us 
00005.389078533 glm0:-> 0x2a WRITE(10) address 00:00, lba 0x004b19d3, len 0x000003, control 0x00 timeout 60 CDBP 300002df0f8 1 sched(0) cdb(10) 2a00004b19d300000300 
00005.824242527 glm0:<- 0x2a WRITE(10) address 00:00, lba 0x004b19d3, len 0x000003, control 0x00 timeout 60 CDBP 300002df0f8, reason 0x0 (COMPLETED) pkt_state 0x1f state 0x0 Success Time 435303us
 ^C    
glm                                                       0            
    value  ------------- Distribution ------------- count              
  4194304 |                                         0                  
  8388608 |                                         3                 
 16777216 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@        678               
 33554432 |@@@@@                                    98                
 67108864 |@                                        27               
134217728 |                                         8                
268435456 |                                         3                
536870912 |                                         0          
#

The implementation of this uses dtrace speculations and as such may require some tuning of the various settings. Specifically the number of speculation buffers I have set to 1000 which should be enough for all but the busiest systems but if you do reach that limit you can increase them using the following options:

-D NSPEC=N

Set the number of speculations to this value.

-D SPECSIZE=N

Set the size of the speculaton buffer. This should be 200 times the size of NSPEC.

-D CLEANRATE=N

Specify the clean rate.

As usual the script is available here. It is version 1.18.

Advertisements

From → Solaris

Leave a Comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: