[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-xfs
Subject:    Re: TAKE - Delay Buffer Changes
From:       Rajagopal Ananthanarayanan <ananth () sgi ! com>
Date:       2001-02-28 16:31:59
[Download RAW message or body]

Danny wrote:
	[ ... ]
> too, and this "hang" may have been due to that.  I've NOT seen the
> "0-order allocation failed" messages, which was my main sticking point.

Memory balancing is handled much better now that XFS plays
"by the rules" of Linux using standard IO paths & daemons.
It's still possible to get allocation failures on HIGHMEM (> 900M memory)
systems on allocations of bounce buffers ... again this is
a problem in linux HIGHMEM IO, there are some discussions on
linux-kernel about reserving a pool of lowmem to allocate for bouncing.

> 
>         Also, either because I'm using the "-o logbufs=4,logbsize=32768" mount
> option, or the recent changes, or both, the Bonnie block writes went
> from 14+ MB/sec to 20+, and the block reads went from 9+ MB/sec to 10+.

Any sequential IO changes are not likely due to log changes.
After the delay buffer change I'm seeing close to raw I/O
speeds on my scsi disk on sequential writes (reads were already
close to optimal).

thanks,

ananth.

--------------------------------------------------------------------------
Rajagopal Ananthanarayanan ("ananth")
Member Technical Staff, SGI.
--------------------------------------------------------------------------

[prev in list] [next in list] [prev in thread] [next in thread]