hp-ux and Disk Read times on DMX-4 Disks

Folks, hope someone might be able to help me out here. I have 2 hp-ux systems on 2 different DMX-4 arrays (DR between 2 sites.)

When I run a 'dd' command on Host_1 I get a time to read the records as about 11 seconds. When I run again I get <1 second, which says to me that there is some caching going on.

When I run the same 'dd' command on Host_2 I get a time to read about 11 seconds but when I run again I still get 11 seconds which I'm reading (perhaps incorrectly) as no useful caching going on.

The above is with a block size of 256k. When I reduce the block size to 128k on the hosts I get for Host_1 read time of about 4 seconds (same count as above) and then a fraction of a second on a re-run. When I run this on Host_2 with the reduced block size of 128k I get the same response asa Host_1 (ie 4 seconds first run, then fraction of second for subsequent runs)

Both of these are on a DMX-4 (different sites) with the same file system layout. Does anyone know of anything I should be looking at to see why Host_2 doesn't behave the same as Host_1 with a block size of 256k in the 'dd' command?

The 'dd' command is ( time dd if=/dev/dsk/c15t15d7 of=/dev/null bs=128k count=1000)

- Hope that all makes sense!

Thanks

Jon

Responses(2)

manny6

9 Posts

0

December 17th, 2009 11:00

no dmx4, but you might want to checkout symstat for cache stats:

symstat example:

/usr/symcli/bin/symstat -sid XX -type requests -dir XX -i 10 -c 2 -port X

Dec17-09:51

DIRECTOR IO/sec Cache Requests/sec % RW

READ WRITE RW Hits

09:50:11 FA-4D 384 353 47 400 63

09:50:22 FA-6A 186 156 41 198 57

09:50:33 FA-6B 219 159 70 230 67

09:50:44 FA-7D 308 214 114 329 67

09:50:54 FA-10D 55 31 27 59 81

09:51:05 FA-11A 172 120 63 183 69

09:51:16 FA-11B 340 235 113 349 66

09:51:26 FA-13D 112 116 6 122 41

PedalHarder

465 Posts

0

December 17th, 2009 17:00

Hi Jon,

Caching is not optional on a DMX. There could be a number of reasons for this behavior... One explanation for the differences in your results is the different workloads and configurations on the 2 arrays. It's not clear in your post if it is the R1 or the R2 side that gets the different results with the block size changes.

Do you have Performance Manager data available to you? One key metric that can help determine what is going on is the "Average LRU Fall Through" time. This is listed against the DA directors. The value is the AVERAGE amount of time data is staying in cache.

As your testing regime is over a very small period of time, you may want to re-test and collect 2 minute REVOLVING performance data to see the differences in the LRU fall through times. The smaller the number the shorter the period of time data is staying in cache. What is the time differenece between the 2 runs of your DD test?

Are the RAID types the same on the R1 and R2 testing devices?

Regards

Jason

View All

No Events found!

Host Systems

hp-ux and Disk Read times on DMX-4 Disks