Subject: Re: Extremely slow IO response times + panic on 3.0_BETA
To: Michael van Elst <firstname.lastname@example.org>
From: Dan LaBell <email@example.com>
Date: 07/14/2005 13:47:53
On Jul 14, 2005, at 1:50 AM, Michael van Elst wrote:
> firstname.lastname@example.org (Dan LaBell) writes:
>> but I decided to test it out on my pc first. It seemed completely
>> zeroed (fresh) , but
>> to verify that ,and to do a read-test, I did a: dd if=/dev/rwd1d
>> bs=128k | sum
>> Afterwards, there would be about a 3 second window, where I could
>> maybe switch vt's,
>> but, then a long hang, which seemed last around a minute, possibly
>> longer, before any kind of responsiveness was returned. I'm wondering
>> if the affect could be tuned away?
> This must be unrelated, since you are reading from a raw device. No
> filecache activity is involved.
> I guess you fall over the end of rwd1d. The size of the raw device
> does not reflect the size of the disk, so it is possible to read
> beyond the end and generate I/O errors that are then naively retried.
Similiar to trying to read from an empty floppy drive? But with no
errors, to console?
And dd reporting no errors, and reading the correct size. I find this
hard to believe.
Also, if I wait 30 seconds or so, and then Ctrl-c I also get a hang,
but not so long.
leading to the hypothesis of some kind of vm tuning issue.
I've never needed to specify size (count) , with using any device file,
when reading it just works, and when writing, perhaps a simple "short
write" and dd reports NNNN + 1 etc, unless of course I want to operate
on the less than the full size.
In fact, I'm so used to this behavior on unix, especially when zeroing
a device, that when I
zero a file, I sometimes forget to add count=