Dave Howorth wrote:
Linda Walsh wrote:
No such thing in standard linux. The cpu nice doesn't affect
the disk-io priority unless you have the non-standard "cfq" scheduling
algorithm enabled. The default when I installed 10.2 recently, I believe
was the 'anticipatory' deadline. Unfortunately, while it may be good for
server workloads, and better for throughput, 'cfq' is better for
interactive use. A background process can easily saturate the disk if
it runs at full speed (even if process is 'niced' down).
I'm curious and didn't find an answer with a quick google. How does one
inspect what algorithm is in use and/or change it? Is it a runtime
option or build-time?
---
It has to be built in when the kernel is built, but I think both
10.2 and 10.3 have the alternate I/O schedulers compiled in as
modules. The default can be set at compile time. So far, I only
know of a way to change schedulers at boot time.
Booting the kernel with the "elevator=" argument allows
specifying an I/O scheduler. Valid values (if they are all
compiled in) are:
elevator=as
elevator=cfq
elevator=deadline
elevator=noop
They choose/enable the "Anticipatory", "completely fair queuing",
"deadline", or "noop" schedulers, respectively. I believe the default
used to be "anticipatory", but this isn't ideal for desktop use, but
tries to optimize total throughput (good for servers processing large
non-interactive programs).
The "ionice" command in util-linux allows modifying cfq
priorities.
Recommended google search terms (from book "Linux Kernel Primer")
would be "Modular IO Schedulers" and "Jens Axboe". In my quick search
just now, didn't see anything about run-time selection, but didn't
go far down the page. The boot time option might be the only way to
go right now...
Sorry couldn't be more help....
linda
--
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
