On Tue, 17 Dec 2002 19:50:04 -0200 (BRST)
Rik van Riel [off-list ref] wrote:

On Mon, 16 Dec 2002, Georg Nikodym wrote:

quoted

Incidentally, a colleague claimed to have seem this behaviour on a
non-rmap 2.4.20.

quoted

1. Known behaviour?
2. Is there any data that I should be collecting that people are
   interested in?
3. Or should I just go back to 2.4.19-rmap14b (which did not trouble
me
   in this way)?

The suspect is the disk elevator, which isn't scheduling requests
in a way to cause lower read latency, but is optimised more for
throughput.  This results in some pauses.

I'll need to look into it.

I discovered after sending the above:

Dec 16 15:08:04 keller kernel: ieee1394: sbp2: sbp2util_allocate_request_packet 
- no packets available!
Dec 16 15:08:04 keller kernel: ieee1394: sbp2: sbp2util_allocate_write_request_p
acket failed
Dec 16 15:08:34 keller kernel: ieee1394: sbp2: aborting sbp2 command

These messages correspond with the pauses...  However, the ieee1394 code
has not changed in some time (as in many months).

-g