BIOLM , DIOLM question

[Wim Van den Wyngaert]

I wrote a script to find the number of outstanding IO's each process has (BIO and DIO).

I noticed that 1 Sybase server had 9 dio and 10 bio outstanding all the time. All 8 other Sybase servers have 1 dio and 2 bio as their base.

Sometimes it increases when there is activity but it never goes below these values.

How is that possible ?

Wim

[Hein van den Heuvel · Accepted solution]

Those processes probably have an outstanding IO for control purposes.
Probably BIO to mailboxes and DIO to TCP sockets.
Check with ANAL/SYS.. SET PROC... SHOW PROC/CHAN and I expect to see 'busy' indicators on the channels with active IO.

Hein.

[Robert_Boyd]

think about it ... you have 8 other processes -- extra dios and bios might be used for synchronization/communication mechanisms with them.


Robert

[Wim Van den Wyngaert]

Yes that must be it.
Now at 8:00 AM I have more activity and the other db servers have significant values too.

Each tcp client has 1 outstanding bio and 1 outstanding dio. Except 1 or 2. I guess these must be dying or doing some keepalive or so (2 have no ISCONNECTED status on socket).

Wim

[Wim Van den Wyngaert]

Another one.

DNS$ADVER has 104 outstanding bio.

It has 4 MB, 8 EI and 2 FW channels open.

And another one.

DTSS$CLERK has 51 outstanding bio.
Is has 1 MB, 3 EI and 1 FW channels open.

How can this be explained ?

Wim

[Wim Van den Wyngaert]

Do they have that many control IO's ?

Wim

[Wim Van den Wyngaert]

Why are control IO's outstanding ? If you do a control IO to a device or whatever, and it never completes, how do you know that the IO didn't fail ?

Dammed difficult to find how many IO's a Sybase server has outstanding to disk ...

Wim

[Ian Miller.]

Wim,
There does not seem to be a good way of looking at the pending I/O queue of a device especially if you are interested I/O requests for a specific process. So I am contemplating writing a program to do this.

At present the current development version of my PWAIT SDA extension will display one of the IRP queues for SCSI&FC disks. If you want to play it is here
http://eisner.encompasserve.org/~miller/pwait.zip

when I get the time will refine this. I am thinking this could be a seperate SDA extension to look at I/O queues.

Any thoughts or opinions on how useful or otherwise such a program would be or what it should do then let me know here or via the email address on http://www.encompasserve.org/~miller/

[Robert_Boyd]

I would guess that most of the outstanding IO for DTS processes are connected with timers and network IO. Some services like that will queue up timers well in advance for many events.

I'm wondering if the usage you're seeing is associated with scheduled events?

Robert