• Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 607
  • Last Modified:

Unbuffered NFS on FreeBSD

I have a process running on a Solaris 2.6 server that writes continuously to a log file. This log file is sitting on an NFS filesystem (hosted by a network application storage filer) which is mounted read-only on a FreeBSD server.

There's a process running on the FreeBSD box that's supposed to read each line as it comes in and add some information about it to a berkely DB file stored on the machine's local disk. Ideally, I'd like the entire process (writting to log -> read log -> add to database) to take a few seconds at most. It currently takes around a minute.

If I tail -f the log file on the Solaris box, I can see the writes happening live, as they happen. If I tail -f the same log file on the FreeBSD box, I see nothing new happening....then about a minute later my console gets flooded with new entries....then nothing for a minute.....then another influx of data.

I think what's happening here is somewhere, the NFS traffic is being buffered by the OS. The only mount options I'm using on the FreeBSD box in /etc/fstab for this partition is 'ro' (read-only). Is there anything else I can do to ensure that the OS itself is not buffering the incomfing NFS traffic? Any sysctl switches that I can tweak?

Thanks for any help you can provide.
0
seagu11
Asked:
seagu11
  • 2
  • 2
1 Solution
 
seagu11Author Commented:
Geez. Sorry, I meant to say the NFS server is a "Network Appliance" filer.  (Netapp). Not a Network Application filer....
Fingers went on autopilot there I guess. :)
0
 
gheistCommented:
most likely you will tweak -r and -w parameters for mount_nfs and maybe noac for nfs misses new attributes, and adjust -a parameter according to your filesizes and reads - the larger -a the larger sequential preread, the worse database and small file performance.

i.e in /etc/fstab:
.. .. .. ro,-r32768,-w32768,noac,-a0
0
 
gheistCommented:
manual for mount_nfs is very handy:

BUGS
     Due to the way that Sun RPC is implemented on top of UDP (unreliable
     datagram) transport, tuning such mounts is really a black art that can
     only be expected to have limited success.      For clients mounting servers
     that are not on the same LAN cable or that tend to be overloaded, TCP
     transport is strongly recommended, but unfortunately this is restricted
     to mostly 4.4BSD servers.
0
 
seagu11Author Commented:
THanks for your pointers.

This actually ended up being resolved by properly synching the time between all involved servers and the filer. They weren't using the same ntp host (and some weren't doing ntp at all).

Sillyness.

Guess I was barking up the wrong tree.
0

Featured Post

Concerto Cloud for Software Providers & ISVs

Can Concerto Cloud Services help you focus on evolving your application offerings, while delivering the best cloud experience to your customers? From DevOps to revenue models and customer support, the answer is yes!

Learn how Concerto can help you.

  • 2
  • 2
Tackle projects and never again get stuck behind a technical roadblock.
Join Now