[Orca-users] Re: Orca NFS Client Graphs Empty

[Waltner, Steve]

I am also seeing the lack of NFS statistics on my systems. The config is SE
Toolkit 3.2.1 collecting data on Solaris 8 using orcallator.se 1.32 and
Solaris 2.6 using orcallator.se 1.27.

I'm seeing basically no NFS traffic reported by Orca, even though these
systems do lots of NFS traffic (one of them is our primary NFS server). My
primary systems use NFS over TCP. I haven't verified that switching to UDP
gives NFS stats, since these are production systems. I had ignored this
omission because I wasn't particularly interested in NFS stats on my
systems. It does appear to be a generic problem with SE Toolkit collecting
data on NFS using the orcallator.se config files.

Steve

> Many thanks for the suggestions, Blair. I had been counting the 
> columns by hand and had concluded that no NFS client activity was 
> being recorded with the default mount options but this was confirmed 
> by running orcallator_column as you suggested. By contrast, there was 
> plenty of NFS server activity logged on the NFS server during the 
> same time period - so the NFS server graphing works for us. 
> Basically, I ran a script that copied some files over NFS on a test 
> system. The following data points show where I had mounted the NFS 
> filesystems with -o proto=udp, which seems to work for NFS client 
> graphs (i.e., we get good graphs generated by Orca):
> 
> orcallator-2002-02-01-000  05:55:00      2.99
> orcallator-2002-02-01-000  06:00:00      7.79
> orcallator-2002-02-01-000  06:05:00     54.97
> orcallator-2002-02-01-000  06:10:00     56.18
> orcallator-2002-02-01-000  06:15:01     12.89
>                            locltime     nfs_call/s
> 
> Then I remounted the NFS filesystems with no options (i.e., default 
> of proto=tcp under Solaris 8) and reran my script. As you can see, 
> zero activity and no graphs (even though nfsstat -c shows plenty of 
> connection-oriented rather than connectionless RPC and the NFS server 
> is graphing plenty of NFS activity):
> 
> orcallator-2002-02-01-000  06:20:00      0.00
> orcallator-2002-02-01-000  06:25:00      0.00
> orcallator-2002-02-01-000  06:30:00      0.00
>                            locltime      nfs_call/s
> 
> I got similar output on a test system running out-of-the-box Orca 
> with the 1.28 configuration files (so percol-* rather than orcallator-
> *).
> 
> Any feedback would be most appreciated. I wonder whether this could 
> be replicated on other systems running Solaris 8 and RICHPse 3.2.1 
> but if no one else out there has seen this before, there must be some 
> unique factor with our systems that I'm missing.
> Cheers,
> Jason
> 
> --- In orca-users at y..., Blair Zajac <blair at o...> wrote:
> > jantevil wrote:
> > > 
> > > Hello --
> > > Orca has been a wonderful tool for us, but we do have one problem
> > > that I wanted to post to the group in case anyone can help. We
> > > noticed that the NFS client graphs were empty even tho' the NFS
> > > server graphs showed plenty of NFS activity (and in fact there is
> > > plenty of NFS activity). After lots of digging, I think I've 
> narrowed
> > > it down to the fact that we use the default of NFS over TCP; when 
> I
> > > remount filesystems with "-o proto=udp", nfsstat -c will show
> > > connectionless rather than connection-oriented client RPC and
> > > suddenly the Orca NFS client graphs start to graph lots of NFS
> > > activity.
> > > Needless to say, we don't want to back down to NFS over UDP just 
> to
> > > get the Orca NFS client graphs, but I have to think that someone 
> has
> > > encountered this issue before and may have some solution. I've
> > > searched the mailing list without any luck, however.
> > > Any suggestions would be appreciated.
> > > Thanks,
> > > Jason
> > 
> > Jason,
> > 
> > The first thing to check is if the output percol-* or orcallator-* 
> files
> > contain NFS data.  If they do, then it's an issue with the Orca 
> configuration
> > files, otherwise it's an issue with orcallator and/or SE.
> > 
> > Check for the nfs_call/s, nfs_timo/s, nfs_badx/s, nfss_calls, 
> v2reads,
> > v2writes, v3reads, v3writes and nfss_bad columns.  You can use
> > orcallator_column.pl to pick a particular column out of your data 
> files
> > like this:
> > 
> >     orcallator_column.pl -c nfs_call/s -c nfss_calls percol-.....
> > 
> > Best,
> > Blair