To: David Marshall <marshall@athena.net.dhis.org>
cc: lkm <linux-kernel@vger.rutgers.edu>
In-Reply-To: <84em66yata.fsf@athena.dhis.org>
MIME-Version: 1.0
Content-Type: TEXT/plain; CHARSET=US-ASCII
Message-Id: <20000610091632.QVJI17505.relay02@chello.nl>
On 9 Jun, David Marshall wrote:
> t.n.vanderleeuw@chello.nl writes:
>
>> I'd say that it appears as if Andrea's classzone patches manage to
>> spread out the nessecary swap-work better than the current VM.
>
> This is stating the obvious, which is why I'm sending it to your
> privately. If you're using IDE drives and haven't used hdparm to
> "unmasq" the IRQ, you might want to try doing so. 'hdparm -u 1
> /dev/hd[whatever]'.
:-)
It's actually not so obvious, I think many people don't know. I've set
both my drives to unmasq the IRQ, increase the multi-sect. count and
other tunings:
/sbin/hdparm -d0 -c1 -m16 -u1 -k1
(for both /dev/hda & /dev/hdc)
This brings me to a point I've been wondering about for the last couple
of weeks:
Without DMA, my drives perform substantially better than with (U)DMA.
The difference is about 1.5MB/sec in throughput as measured with
/sbin/hdparm -T -t
With bonnie++ I observe a very similar difference in performance, with
a relatively small increase in CPU usage. I can post more detailed
stats if wanted.
The chipset is a SiS5513/5597, and UDMA is enabled for both the drive
and the chipset.
So I'm wondering why (U)DMA is so much slower than PIO. Is there anyone
who has a clue?
Thanks for any info!
--Tim :-)
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Jun 15 2000 - 21:00:20 EST