Bob,

I ran a little speed-test to compare nfs and samba, using a single 65M file,
as described in the NFS-Howto.  In summary, the file took 19sec to read off
the local disk, 23sec to read from another host using nfs, and 27sec to read
from another host using samba.  Looks like it's time to upgrade again.

I was surprised that nfs tested faster and uses less cpu time than smb.  The 
server here is not too fast, and I expect the limiting factor here is the 
relatively slow hard drive on "l2".  

Host "fast" is running Mandrake 8.0 on an AMD-K2/350(?)
Host "l2" is running RedHat 6.1 on an AMD-K2/233, "cheapo" ide hard-drive,
UDMA maybe not working but at least it's still spinning a little?
Network is 100BaseT (netgear/tulip nic's) with little traffic.

Ralph
--clip--
[root@fast root]# mount l2:/home /mnt/l2/home/

[root@fast root]# time dd if=/dev/zero of=/mnt/l2/home/temp bs=16k count=4096
4096+0 records in
4096+0 records out
0.04user 3.49system 0:21.28elapsed 16%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (168major+29minor)pagefaults 0swaps

[root@fast root]# time dd if=/mnt/l2/home/temp of=/dev/null bs=16k
4096+0 records in
4096+0 records out
0.04user 2.42system 0:23.13elapsed 10%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (168major+33minor)pagefaults 0swaps

[root@fast root]# mount -t smb //l2/doc /mnt/l2/usr
Password:

[root@fast root]# time dd if=/dev/zero of=/mnt/l2/usr/temp bs=16k count=4096
4096+0 records in
4096+0 records out
0.07user 4.88system 0:27.23elapsed 18%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (168major+29minor)pagefaults 0swaps

[root@fast root]# time dd if=/mnt/l2/usr/temp of=/dev/null bs=16k
4096+0 records in
4096+0 records out
0.06user 6.01system 0:26.60elapsed 22%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (168major+33minor)pagefaults 0swaps
--clip--

[root@l2 /root]# cat /etc/exports
/home   (rw,no_root_squash)
[root@l2 /root]# time dd if=/home/temp of=/dev/null bs=16k
4096+0 records in
4096+0 records out
0.02user 4.22system 0:18.98elapsed 22%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (16528major+16minor)pagefaults 0swaps
[root@l2 /root]#
--clip--

At 05:48 PM 8/21/2001 -0700, "Bob Crandell" <[EMAIL PROTECTED]> wrote:
>Actually, I was more interested in the transfer between the two Linux
boxes.  I know WinDoze is slow and I don't care.  It's just one more reason ...
>
>Another test was to play an MP3 over an NFS connection.  The music skipped
and paused.
>
>I was wondering if there was a way to tune NFS.  You mentioned rsize and
wsize, now I have a place to start checking.
>
>Thanks
>
>>>> [EMAIL PROTECTED] 8/21/2001 2:42:18 PM >>>
>Bob,
>
>I'm pretty clueless about nfs, because samba works well enough for me and
>it doesn't require additional stuff on the clients, but here are my ideas
>anyway...
>
>1) Did you ask the LabF people about the slowness?  Is it only slow hooking
>to a linux server, or is it slow on LabF servers too?
>
>2) nfs is very sensitive to the rsize and wsize options.  I'm not sure what
>the defaults are, but I used 8k when I was trying to optimize on a 10base2
>network.  Maybe the LabF nfsclient is using odd block sizes that are 
>optimized for their own nfs-server?
>
>Ralph
>
>At 01:28 PM 8/21/2001 -0700, "Bob Crandell" <[EMAIL PROTECTED]> wrote:
>>Larry,
>>
>>There are no error messages.  With my setup, I'm assuming it stays mounting
>all the time.  This is how I responded to Randolph:
>>There are 2 Linux boxes and 1 Windows box.  My test is a directory with a
>little over 5 M of various size files copied between the 2 Linux boxes using
>either Windows or Linux.  Linux is, understandably, faster but it is still
>slow.  Windoze took 15 minutes to do this copy.
>>
>>Windoze is using nfsAxe from http://LabF.com.  This allows Windoze to be a
>nfs client or server.
>>
>>Both /etc/exports files look like this:
>>/home  192.168.1.0/22(rw,no_root_squash)
>>
>>One /etc/fstab looks like this:
>>oci:/home  /mnt  nfs  rw,noauto,user  1  1
>>
>>The other looks like this:
>>server:/home  /mnt  nfs  rw,noauto,user  1  1
>>
>>Thanks
>>
>>>>> larry a price <[EMAIL PROTECTED]> 8/21/2001 11:57:11 AM >>>
>>On Tue, 21 Aug 2001, Bob Crandell wrote:
>>
>>> Hiring Larry?
>>
>>I was actually more interested in what you were up to though,
>>if there is work for which m talents are suited I am available subject to 
>>some strictures. <---(can't compromise the health insurance)
>>
>>> Depends on 2 things;  The answer to the NFS slowness question and
>acquiring a client or two.
>>>
>>I don't have access to the setup you're having difficulties with so in
>>this case debugging advice is about psychic friends quality.
>>
>>Is it mounting and unmounting the share for each transaction? That would
>>slow it down some. Have you tried intrumenting various parts of the sstem?
>>what error messg have you gotten?
>>
>> 
>>> >>> [EMAIL PROTECTED] 8/21/2001 4:26:33 AM >>>
>>> On Sun, 19 Aug 2001, Bob Crandell wrote:
>>> 
>>> > Soon to be <Company Name>
>>> Oh Cool? Is <Company name=$name terms=$contract valuta=$somefigure>
>>> HIRING?
>>> 
>>> Enquiring Minds Want to know ;)
>>> > 
>>> 
>>> me thinks the above tuple actually belongs to either the <bid> or  <ask>
>>> tags for interactive use. quel pensee avez vous au sujet de la?
>>> 
>>> 
>>
>

Reply via email to