On Sun, 1 Apr 2001, Ralston, Steve wrote:

>       > I get a reasonable number of 'ordered tag forced' messages. As in
>       > Apr  2 11:20:34 swtf kernel: sym53c895-0-<3,0>: ordered tag
> forced.
>       > This doesn't appear to have any adverse effect on the filesystem
> ..,
>       > it's just that I'm curious as to why it is/would happen.
> 
> Refer to comment:
>   /*
>   **  Force ordered tag if necessary to avoid timeouts
>   **  and preserve interactivity.
>   */
> and following block of code near line 6619 in 2.4.3'ish
> drivers/scsi/sym53c8xx.c.
> Looks like this particular scsi driver throws out ORDERED_TAG every
> 3 seconds or so...

Almost true. :)

The driver will not fire any ordered tag if all IOs complete within 3
seconds (3 seconds delay hard-coded). In fact the driver uses 2 counters
of tagged commands. When switching to the other counter (mostly every 3
seconds), if this counter is not zero then the value represents the number
of tagged commands that haven't completed within at least 3 seconds (and
at most something lower than 6 seconds).

The message may well be harmless. But since it is printed out with
verbosity=1, the syslog might be flooded a bit. User can hack the driver
for this message to be printed out for higher verbosity, or ensure
verbosity is minimal (Verbose mode not set in NVRAM and boot with
sym53c8xx=verb:0).

> Many scsi drivers inject intermittant ORDERED_TAG like this to keep
> certain SCSI targets from starving WRITE IOs under heavy load.
> Hee, I spent time as "guest" (more like prisoner) of Hitachi in Toyko many
> years
> ago due to a problem of this nature.  They were using IBM fast+wide drives
> which had a tendency to defer/starve WRITE IOs for too long under heavy
> load.

Indeed. My first SCSI disk was an IBM S12 that starved tagged commands a
lot. But I have seen recent hard disks reordering IOs heavily too
(hundreds of IOs being passed). But they are so fast than the 3 seconds
delay is unlikely to be hit very often. This does not apply to raid
devices that are slow by design... :-)

Regards,
  Gérard.

-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]

Reply via email to