[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: NFS recommendation
On Fri, 04 Jun 1999 21:27:03 +0300 (IDDT), Ariel Biener
<ariel@fireball.tau.ac.il> wrote:
>On Fri, 4 Jun 1999, Udi Finkelstein wrote:
>
>
> Hi,
>
>
> Here is what I have to offer.
>
>
> From my experience, it would be a "BAD" idea to tie up between the
>system partitions and this partition you want to make available to Solaris
>clients.
Obviously, the data to be shared will be on a different partition.
> So, for starters, get a fast (7200 rpm) ATA disk for the system,
>inside the computer.
The 22GB disks are 7200RPM, and we preferred them over the 25GB 5400RPM disks,
which cost the same.
> Install RedHat 6.0, not 5.2. Pay attention that the 2.2.x kernels have
>a much more aggressive approach towards NFS. This broke some clients that
The question is, is RH5.2 that bad? Is RH6.0 stable enough?
If performance goes up from 6MB/s to 8MB/s, I'de stay with 6MB/s, if it's
more stable. If performance is 0.6MB/s, that's a different story...
>didn't implement NFS correctly, like NetApp OS (OnTAP). Network Appliance
>have fixed it (5.2.1P2 is fixed).
What has the NetApp got to do with this? I thought it was only an NFS server.
> If you chose 2.2.x kernels, do not use knfsd yet. It is still in
>development, and it's not a wise choice till it's stable, since I guess
>you want to provide a stable environment. Get the latest nfsd server
>(version 2.2beta40 or newer).
thanks. Which kernel should I use? should I take 2.2.9, or stick with the
2.2.5 distributed with RH6.0?
> Also, the good stuff about the 2.2beta40 (I use beta43) rpc.nfsd is
>that it can finally run a few servers simulatenously, without disabling
>(rw) like before. This should improve performance.
Where can I get it? a freshmeat search for nfsd proved no results.
> If the answer to this is that there are 20 clients or more, each
>writing and reading all the time (and not the same files over and over,
>which are cached), and at 100Mbit per workstation, you might think of
>getting a Gigabit switch, where the NFS server will work at 1Gigabit rate
>to the switch, and the clients at 100Mbit/s. But this only if performance
>is critical. You should then use the ATA/66 card, to improve disk IO
>performance.
Even without a GigaBit switch, things will be much better than now, as even
our NetApp is still on a single 100MBit/s interface.
As for an ATA/66 card, we may get one later, but only because we may need more
disks. AFAIK, it already has a Linux driver.
> To conclude, SGI is gonna release XFS, and ext3fs file system is also
>around the corner. These file systems support striping in a more native
>and stable manner, and both ofer Journaling.
Yes, but this will take a long time. Hopefully I won't be needing a single
file >2GB until these are ready ... (my typical files at the moment are
~150MB).
thanks,
Udi