lkml.org 
[lkml]   [2000]   [Aug]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Can't get a block out edgewise
Thanks for your response.  Actually the dd command is only a simplified
version of the real problem. It appears that any big I/O operation slugs the
system. The original problem was big SMB file writes over a 100MB/s link.

I have since found a tuning that seems to keep the system going fairly well.
I'm using:

echo "40 100 512 512 2000 3000 5000 1884 2" > /proc/sys/vm/bdflush

--
jeffrey hundstad



Ricky Beam wrote:

> On Thu, 3 Aug 2000, Jeffrey E. Hundstad wrote:
> ...
> >I open two sessions. In one I say "find /", in the other I say "dd
> >if=/dev/zero of=big.file bs=10240K count=3".
>
> Umm, "block size == 10MEG"?!?! Well, right there is half your problem.
> It's going to take some time to 10M zeroed buffer and then handing 10M
> in one shot to the driver... the system freezes during the 10M read()
> of /dev/zero. It becomes very unresponsive during buffer flushing as
> a side effect of the way SCSI command queuing works. (Once the driver
> command queue is full, no more commands will be sent to the driver until
> the queue is empty. I _HATE_ that -- esp. with a DPT I2O controller that
> takes _210_ commands and can take _5 MINUTES_ to execute all of them.)
>
> >Is there any way to smooth out the performance a bit? I'd like some
> >writes to occur then some reads so file I/O doesn't grind to a halt.
>
> change the bs=## to something smaller. The largest block size the SCSI
> layer will handle is 64k (it can theoretically go much higher, but doesn't)
>
> --Ricky


-
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/

\
 
 \ /
  Last update: 2005-03-22 13:58    [W:0.234 / U:0.164 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site