NFS linux->osf1 slow, linux->irix fast

NFS linux->osf1 slow, linux->irix fast

Post by Remco Wout » Fri, 22 Jan 1999 04:00:00



Hi,

We have a mixed network of machines running linux, osf1, and irix. All
machines serve disks to each other over NFS. There seems to be a huge
performance problem when writing on linux to disks mounted from our
ALPHA'S, whereas writing to the SGI's works as advertised. I did a
dejanews search and a few people mention the same problem but no
solution has been posted. Usually linux get's the blame. Since
everything works fine with IRIX I really think there is something
funny with the NFS2 implementation under OSF1.

Attached is the output of tcpdump for writing to files on disk on two
of our machines, everything is on 100Mb ethernet half duplex.  The
problem seems to be that under OSF1 writes are serialized and under
IRIX they are not. The same thing is seen when writing large amount of
blocks.

I hope somebody can shed some light on the situation, so I can try to
fix things.
--

tel/fax:+46-18-4714642/511755    | PO Box 576, S-751 23 Uppsala, Sweden.
 WinErr: 006 Malicious error - Desqview found on drive

TCPDUMP
-------
Linux rembrandt.bmc.uu.se 2.0.36
IRIX bosch 6.3 12161207 IP32
OSF1 carr.bmc.uu.se V4.0 564.32 alpha

dd if=/dev/zero of=<file on carr (OSF1)>  bs=1024 count=5

09:51:46.643697 rembrandt.1427250621 > carr.nfs: 152 lookup [|nfs]
09:51:46.643697 carr.nfs > rembrandt.1427250621: reply ok 128 lookup [|nfs]
09:51:46.643697 rembrandt.1427250622 > carr.nfs: 148 lookup [|nfs]
09:51:46.643697 carr.nfs > rembrandt.1427250622: reply ok 128 lookup [|nfs]
09:51:46.643697 rembrandt.1427250623 > carr.nfs: 172 setattr [|nfs]
09:51:46.663697 carr.nfs > rembrandt.1427250623: reply ok 96 setattr [|nfs]
09:51:46.663697 rembrandt.1427250624 > carr.nfs: 1180 write [|nfs]
09:51:46.693697 carr.nfs > rembrandt.1427250624: reply ok 96 write [|nfs]
09:51:46.693697 rembrandt.1427250625 > carr.nfs: 1180 write [|nfs]
09:51:46.743697 carr.nfs > rembrandt.1427250625: reply ok 96 write [|nfs]
09:51:46.743697 rembrandt.1427250626 > carr.nfs: 1180 write [|nfs]
09:51:46.803697 carr.nfs > rembrandt.1427250626: reply ok 96 write [|nfs]
09:51:46.803697 rembrandt.1427250627 > carr.nfs: 1180 write [|nfs]
09:51:46.863697 carr.nfs > rembrandt.1427250627: reply ok 96 write [|nfs]
09:51:46.863697 rembrandt.1427250628 > carr.nfs: 1180 write [|nfs]
09:51:46.913697 carr.nfs > rembrandt.1427250628: reply ok 96 write [|nfs]

dd if=/dev/zero of=<file on bosch(IRIX)>  bs=1024 count=5

09:51:51.273697 rembrandt.1427250629 > bosch.nfs: 152 lookup [|nfs]
09:51:51.273697 bosch.nfs > rembrandt.1427250629: reply ok 128 lookup [|nfs]
09:51:51.273697 rembrandt.1427250630 > bosch.nfs: 148 lookup [|nfs]
09:51:51.273697 bosch.nfs > rembrandt.1427250630: reply ok 128 lookup [|nfs]
09:51:51.273697 rembrandt.1427250631 > bosch.nfs: 172 setattr [|nfs]
09:51:51.293697 bosch.nfs > rembrandt.1427250631: reply ok 96 setattr [|nfs]
09:51:51.293697 rembrandt.1427250632 > bosch.nfs: 1180 write [|nfs]
09:51:51.293697 bosch.nfs > rembrandt.1427250632: reply ok 96 write [|nfs]
09:51:51.293697 rembrandt.1427250633 > bosch.nfs: 1180 write [|nfs]
09:51:51.293697 bosch.nfs > rembrandt.1427250633: reply ok 96 write [|nfs]
09:51:51.293697 rembrandt.1427250634 > bosch.nfs: 1180 write [|nfs]
09:51:51.293697 bosch.nfs > rembrandt.1427250634: reply ok 96 write [|nfs]
09:51:51.293697 rembrandt.1427250635 > bosch.nfs: 1180 write [|nfs]
09:51:51.293697 bosch.nfs > rembrandt.1427250635: reply ok 96 write [|nfs]
09:51:51.293697 rembrandt.1427250636 > bosch.nfs: 1180 write [|nfs]
09:51:51.293697 bosch.nfs > rembrandt.1427250636: reply ok 96 write [|nfs]

 
 
 

NFS linux->osf1 slow, linux->irix fast

Post by Torsten Blan » Tue, 26 Jan 1999 04:00:00


... sorry ... no solution.

Here we have the same problem. I have posted in lots of newsgroups, and
asked lots of linux people, without any success. The nfs data flow
from the Linux PCs to the OSF DEC Alphas is limited to 250 kB/s
independent of the band width. Neither installation of the new 2.2.0
kernel nor playing with the mount options wsize/rsize does help.

  Torsten Blank

 
 
 

1. FTP problem: linux>>win/SGI slow win/SGI>>linux fast

The machine is on a Novell 4.2 network and uses an Intel
Ether Express 16 card. It is possible to be on another
machine and transfer a file FROM the linux box at a rate of
about 600kB/s.
However, transferring to the linux machine gives rates of
about 18kB/s.  Similarly if one is
on the Linux box using ftp, transfer both from and
to another machine and
from another machine is slow - around 18Kb / sec.

It is a dual boot system - when tranferring/receiving files
using windoze ftp (QVFTP or WS_FTP) the transfer rates in
both direction are 600kB/s.

At installation time, linux was set up to use the
'eexpress' driver and the bootp protocol.

There are no packet overflows, packets lost or errors generally.

Visualising the transfer using the 'hash' function of ftp
shows that transfer progresses at the expected speed for
10-30kB, then stalls or slows down to a snail pace, then
starts again, slow, stall, slow, etc...

I have tried
- explicitly stating the IRQ=10, I/O base address 0x300
- turning on/off IPX
- making the MTU smaller
- manually configuring (i.e no bootp) the ip address etc.
- playing with the routing window as suggested in the
HOWTO-Ethernet file.

The only effect I have been able to achieve is a reduction
in performance to 13kB/s.

Checking usenet a number of people seem to have had a
similar problem but no relevant solutions suggested.  One
person claimed they started having this problem when
they upgraded their motherboard.

I have no idea what to do next... help...

2. Disk Mirroring Under Solaris 2.4x86

3. Slow Network Linux->Win98=Slow Win98->Linux=Fast

4. automated ftp

5. FTP problem: Linux->Win/SGI slow Win/SGI->Linux fast

6. Can't ls or dir in ftp

7. 95 -> linux fast; but linux -> 95 slow!

8. Very Slow push during NIS make

9. 'doze -> linux fast, linux -> 'doze slow

10. Network fast Linux -> WinXP but slow WinXP -> Linux

11. rpc.mountd: getfh failed: Operation not permitted ( WinNT -> samba -> Linux -> NFS -> Solaris)

12. Extremely slow win95->ethernet->linux->(modem PPP) connection.

13. WPO->NWSunLink->Solaris1.1->NFS->Solaris2.3