>...
>
>On Fri, Jun 03, 2005 at 09:26:07PM -0400, Thor Lancelot Simon wrote:
>> On Fri, Jun 03, 2005 at 01:30:26PM -0400, Sean Davis wrote:
>> > 
>> > Are you running with softdeps enabled, too? I've had softdeps and
>> > new_bufq_strategy bring this system to it's knees when doing a lot of stuff
>> > at once - but running without softdeps, without new_bufq_strategy,
>> > everything works just fine.
>> 
>> I can't reproduce what you're describing, but I note that you seem to be
>> supporting a claim about the effect of one variable by using evidence in
>> which you explicitly admit that it is confounded with another variable.
>
>I've had it happen with just softdep, and with just new_bufq. The effects
>are more severe with just new_bufq.
>
>> That is not very persuasive to me.
>
>I honestly couldn't care less if I "persuade" you. I'm just saying that
>softdep & new_bufq are not wins on my system. Whether or not it changes your
>worldview is up to you.
>
>-Sean
>
	Just another data point:  On nearly every machine I have (almost
all i386 or amd64, but a couple of old mac-ppc's) tried NEW_BUFQ_STATEGY.
On most it makes little apparent difference;  But on the many machines I
have running raidframe (RAID5 on IDE - one disk per channel 4 to 9 disks
per machine), NEW_BUFQ_STATEGY absolutly kills any interactive performance
- independant of softdeps (I've tries both ways - with softdeps enabled and
disabled).  In general on the same machines, softdeps is a *huge* win.

	Please do not make new_bufq the default (it has already been said
that it is expected to hurt performance on IDE drives with write caching
enabled - given large and reliable UPSs, this is certainly not a rare setup,
even for critical production machines).

	Paul Shupak
	track@plectere.com