Conference ssdevo::hsz40_product

784.0. "HSZ performance evaluation tools ?" by FIREBL::LEEDS (From VAXinated to Alphaholic) Thu Feb 27 1997 12:36

One of our largest Digital UNIX production customers is struggling with
performance evaluation and bottlenecks in their configuration. 

Basically, it is three 8400s, with about 12 shared FWD SCSI busses,
connected to 12 HSZ40 pairs, with about 1.3TB of RZxx disks. 

They have tools like PolyCenter Performance Advisor which help them analyze 
CPU and Memory bottlenecks, but they are missing important I/O subsystem 
information.

What they are desperately asking for is a tool to monitor the HSZs. 
Specifically they want answers to the following questions:

 - How do we determine if our read/writeback caches are effective ?
 - How do we determine if upgrading to HSZ50s with more cache would help ?
 - How de we find out if we have any SCSI bottlenecks behind the HSZs ?
 - For the RAID sets, how do we determine if there are any spindle 
   bottlenecks ?
 - For the HSZ redundant pairs, is our load evenly balanced across the pair?
 - Is the FWD bus between the KZPSA and the HSZ40 pair a bottleneck ?

  etc., etc.

They are coming up on their BUSY summer season, and want to make any 
hardware changes necessary prior to their peak season, but they don't know 
how to eveluate the performance of the I/O subsystem.

We must have some tools similar to the old HSC cache evaluation tool which 
will run on an HSZ to help report on cache, bus, controller and spindle
usage/bottlenecks - don't we ???  They can't watch VTDPY for 24 HSZs over a 
5 day period to determine where the bottlenecks are, there must be a better 
way ...

Any help would be appreciated ...

thanks

Arlan

Hi

 Have you tried running dskcoll or collect?

see http://www-unix.zk3.dec.com/tuning/tools/tools.html

and http://www-unix.zk3.dec.com/tuning/tools/sys_check/sys_check.html

I would suggest you want to look at some queue depth data.

Have you run sys_check on the systems...and reviewed the HSZ
setttings?

Another way to look at the HSZ performance is vtdpy from a terminal
connected to the hsz....I don't know what performance data can be
seen from the SWCC.

> Have you tried running dskcoll or collect?
>
>see http://www-unix.zk3.dec.com/tuning/tools/tools.html
>
>and http://www-unix.zk3.dec.com/tuning/tools/sys_check/sys_check.html

nope - didn't even know they existed - neither did the customer...

>Another way to look at the HSZ performance is vtdpy from a terminal
>connected to the hsz....I don't know what performance data can be
>seen from the SWCC.

As stated in .0, VTDPY running on 24 HSZs to obtain meaningful data is not 
realistic.....

Thanks for the pointers - I'll check them out..

Arlan