Big Reads

[Mark Parsons]

Our storage team have noticed that a VPlex cluster is getting a host initiator with big reads. The host initiator is pointing to a hpux server running HP11.31.

I guess that big reads is pointing to block size.

Are there any tools out there with hpux that can help me check this and also reduce the read size request.

[Duncan Edmonstone]

Mark,

I have never heard of anyone having this issue before - and I have never heard of anyone tuning the maximum block size that an FC based initiator is using with HP-UX.

That said, I would start with the scsimgr(1m) command and scsimgr_esdisk(7) attributes and this technical paper:

https://support.hpe.com/hpsc/doc/public/display?sp4ts.oid=5372858&docLocale=en_US&docId=emr_na-c01914672

I can't see any indication that block size is an actual tunable however.

[Bill Hassell]

Reducing read size is definitely going to impact performance on any server. The goal for any database is to get the required data as quickly as possible and read-ahead stacking for small blocks is a common method to improve performance. It would be useful to know what a "big read" means from the storage team.

That being said, it is possible that someone is running a dd command for some unknown purpose which can be configured for multi-megabyte sequential records. I would run glance to find the processes generating the most disk I/O. Then determine if those processes are working as designed.