From: Kenn Humborg (kenn at domain linux.ie)
Date: Sun 28 May 2000 - 16:47:14 IST
On Sun, May 28, 2000 at 04:28:58PM +0100, Thomas Ribbrock wrote:
> On Sat, May 27, 2000 at 02:36:30AM +0100, Niall O Broin wrote:
> > I wouldn't blame your mail downloading slowness on the drive - a don't think
> > your modem is outrunning even the slowest drive, and even if it were,
> > files saved to disk are first saved in RAM so it would be the flusing
> > process that would be slower, rather than your email download.
> I experimented a bit more with hdparm, among other things enabling 32bit
> data transfers (which should be possible, with the IDE interface running of
> the Vesa Local Bus), however to no avail. The IDE chip is a Promise 2020.
Have you tried enabling 'unmask irq'?
# hdparm -u 1 /dev/hda
This makes a huge difference to serial I/O throughput with IDE CD drives.
What this option does is tell the IDE drivers to re-enable interrupts sooner
rather than later when handling an interrupt from the IDE controller.
Without this option, interrupts from the UART driving your modem can
get lost, causing the UART's input buffer to overflow, corrupting the
PPP frames, resulting in TCP retries and, thus, slower downloads.
Your older drive and controller combination, along with the slower
CPU, might be causing this same problem with disks as well as CD
drives. Try pinging your upstream router and doing a find / at the
same time. If the ping is affected, then try the hdparm command above
and see if it improves the ping performance.
[Donncha, this would be a good candidate for an IO link.]
This archive was generated by hypermail 2.1.6 : Thu 06 Feb 2003 - 13:06:17 GMT