fcsmutil queue stats

[Christof Schoeman]

Hi there

I have a question of a theoretical nature - hope you can help.

I'm trying to determine which queues are involved in the I/O subsystem and what their capacities are. I hope I've got the sequence right.

When an I/O reaches the device driver, it enters the "wait queue" which has no limit, as far as I know.

When the I/O is processed by the device driver, it enters the the "active queue". Is there a limit on the number of I/Os that can be "in flight"? If so, what determines this value?

On the HBA level, I found the following statistics:

#fcmsutil /dev/fcd0 stat

reports a "Request Queue Full" value. Does this value indicate the number of I/O requests that the HBA can accept from the device driver? If so, what is the limit? If not, where does this fit into the picture?

#fcmsutil /dev/fcd0 devstat *nport_id*

reports a "Send failure - Request queue was full" value. How does this differ from the request queue reported above? Again, what are the limits?

The above command also reports an "IOs failed due to SCSI queue full" value. It almost sounds if this queue is on a higher level than the request queues above. Where does this queue fit into the picture? What are the limits?

I know these are probably not very practical questions and may even be considered silly, but I would REALLY like to understand the I/O subsystem down to the bits and bytes level. Even a hint as to where I can find some more information, would be appreciated.

[Steven E. Protter]

Shalom,

There is no such thing as an endless queue. The card has limits as to how many requests it can hold. The OS will act as a buffer for i/o requests as well.

This part of your question:
>>>
reports a "Send failure - Request queue was full" value. How does this differ from the request queue reported above? Again, what are the limits?

The above command also reports an "IOs failed due to SCSI queue full" value. It almost sounds if this queue is on a higher level than the request queues above. Where does this queue fit into the picture? What are the limits?
>>>

Indicates to me that your card is running at or near capacity. You may find that there are high i/o wait times. To get that you need glance/gpm or the non-graphical equivalent,
http://www.hpux.ws/system.perf.sh

SEP

[Christof Schoeman]

Thanks SEP

If there is no such thing as an endless queue, what is the limit of the wait queue?

And the active queue?

The values reported by fcmsutil are all 0 for the metrics metioned above, which means that the card has never encountered any of these conditions.

What I'm really looking for, is what the limits of these queues are, and also what the differences are between the request queue for the card (stat), the request queue for the nport (devstat) and the SCSI queue.

Thanks again.