terrible performance with "sync" mount option (needed for subfs)
Hi, my system is running SuSE 9.0, using kernel 2.6.8.1 with the packet writing patches from http://w1.894.telia.com/~u89404340/patches/packet/2.6/ as well as the bugfix patches from the LKML from the thread "Packet writing problems" (August 16th). I also applied the udftools patch for 1.0.0b3. I setup the drive with pktsetup and mounted it with "mount -t udf -o rw,noatime /dev/pktcdvd/mycd /media/udf" and everything works fine. However, what's not working is the "sync" option. When I mount with "-o rw,noatime,sync" and start copying a 49MB file, it takes more than one hour until it's finished. With ls I can indeed monitor the copy progress is about 10kb/s. Is that a bug, or a design problem in the kernel? Note that with the "sync" option working, subfs could perfectly be used for packet writing! I tried it with "mount -t subfs -o fs=udf,rw,noatime dev/pktcdvd/mycd /media/udf" and it works great. You can even access the drive in parallel via /media/udf and /media/cdrecorder (which is setup with subfs, too). But to avoid users removing the cd before its written, the sync option is neccessary with subfs... BTW, copying a 100MB file with dd to the cd and syncing afterwards took 3:45 minutes alltogether with a 4x cd-rw. Is that an acceptable speed? cu, Frank -- Dipl.-Inform. Frank Steiner Web: http://www.bio.ifi.lmu.de/~steiner/ Lehrstuhl f. Bioinformatik Mail: http://www.bio.ifi.lmu.de/~steiner/m/ LMU, Amalienstr. 17 Phone: +49 89 2180-4049 80333 Muenchen, Germany Fax: +49 89 2180-99-4049
Frank, Not actually related to packet writing, but when writing to a firewire hard drive, an operation taking a minute under normal mount conditions takes over 15 minutes with the sync option. Basically, no buffering of writes takes place, so every disk operation has to be run individually. Steve On Wednesday 01 September 2004 15:11, Frank Steiner wrote:
Hi,
my system is running SuSE 9.0, using kernel 2.6.8.1 with the packet writing patches from http://w1.894.telia.com/~u89404340/patches/packet/2.6/ as well as the bugfix patches from the LKML from the thread "Packet writing problems" (August 16th). I also applied the udftools patch for 1.0.0b3.
I setup the drive with pktsetup and mounted it with
"mount -t udf -o rw,noatime /dev/pktcdvd/mycd /media/udf"
and everything works fine. However, what's not working is the "sync" option. When I mount with "-o rw,noatime,sync" and start copying a 49MB file, it takes more than one hour until it's finished. With ls I can indeed monitor the copy progress is about 10kb/s.
Is that a bug, or a design problem in the kernel?
Note that with the "sync" option working, subfs could perfectly be used for packet writing! I tried it with
"mount -t subfs -o fs=udf,rw,noatime dev/pktcdvd/mycd /media/udf"
and it works great. You can even access the drive in parallel via /media/udf and /media/cdrecorder (which is setup with subfs, too). But to avoid users removing the cd before its written, the sync option is neccessary with subfs...
BTW, copying a 100MB file with dd to the cd and syncing afterwards took 3:45 minutes alltogether with a 4x cd-rw. Is that an acceptable speed?
cu, Frank
-- Dipl.-Inform. Frank Steiner Web: http://www.bio.ifi.lmu.de/~steiner/ Lehrstuhl f. Bioinformatik Mail: http://www.bio.ifi.lmu.de/~steiner/m/ LMU, Amalienstr. 17 Phone: +49 89 2180-4049 80333 Muenchen, Germany Fax: +49 89 2180-99-4049
-- Steve Boddy
Stephen Boddy wrote:
Frank,
Not actually related to packet writing, but when writing to a firewire hard drive, an operation taking a minute under normal mount conditions takes over 15 minutes with the sync option. Basically, no buffering of writes takes place, so every disk operation has to be run individually.
Ok, I think I get that. So "sync" means that every single operation is synced. I thought of it more like "do the operation and then call sync", because it looks like that when e.g. writing to a floppy. But it seems that buffering is crucial for packet writing, from what I read in the lkml threads. Hmm, so looks like packet writing is not so suitable for subfs... Thanks! cu, Frank -- Dipl.-Inform. Frank Steiner Web: http://www.bio.ifi.lmu.de/~steiner/ Lehrstuhl f. Bioinformatik Mail: http://www.bio.ifi.lmu.de/~steiner/m/ LMU, Amalienstr. 17 Phone: +49 89 2180-4049 80333 Muenchen, Germany Fax: +49 89 2180-99-4049
On Thu, Sep 02 2004, Frank Steiner wrote:
Stephen Boddy wrote:
Frank,
Not actually related to packet writing, but when writing to a firewire hard drive, an operation taking a minute under normal mount conditions takes over 15 minutes with the sync option. Basically, no buffering of writes takes place, so every disk operation has to be run individually.
Ok, I think I get that. So "sync" means that every single operation is synced. I thought of it more like "do the operation and then call sync", because it looks like that when e.g. writing to a floppy. But it seems that buffering is crucial for packet writing, from what I read in the lkml threads.
Hmm, so looks like packet writing is not so suitable for subfs...
subfs -o sync mount needs to die anyways, it sucks on every device not just packet writing devices. -- Jens Axboe
Jens Axboe wrote:
Hmm, so looks like packet writing is not so suitable for subfs...
subfs -o sync mount needs to die anyways, it sucks on every device not just packet writing devices.
At least for floppies it works quite well here... And indeed, yast2 is setting this as default option for floppies in SuSE 9.1. cu, Frank -- Dipl.-Inform. Frank Steiner Web: http://www.bio.ifi.lmu.de/~steiner/ Lehrstuhl f. Bioinformatik Mail: http://www.bio.ifi.lmu.de/~steiner/m/ LMU, Amalienstr. 17 Phone: +49 89 2180-4049 80333 Muenchen, Germany Fax: +49 89 2180-99-4049 * Rekursion kann man erst verstehen, wenn man Rekursion verstanden hat. *
On Mon, Sep 06 2004, Frank Steiner wrote:
Jens Axboe wrote:
Hmm, so looks like packet writing is not so suitable for subfs...
subfs -o sync mount needs to die anyways, it sucks on every device not just packet writing devices.
At least for floppies it works quite well here... And indeed, yast2 is setting this as default option for floppies in SuSE 9.1.
that's because floppy performance sucks so much already that you don't notice. -- Jens Axboe
participants (3)
-
Frank Steiner -
Jens Axboe -
Stephen Boddy