very chatty clients sent 28June2002

[Tom Wike]

I wanted to share the comments from one of our NIS+ 'experts' on how
RedHat's NIS+ client implementation is affecting our NIS+ servers, if we
get more than a few hundered Linux NIS+ clients this could turn out to 
be a serious problem for our network.

-Tom

On Thu, Oct 31, 2002 at 08:12:34AM -0600, Chris Barrera wrote:
> 
> If the linux box is running as a nis+ client, and using the nis+
> protocol to talk to a nis+ server, where does compat mode on the
> server come into play? I am assuming one means NIS compat mode here.
> Even assuming the NIS (not +) clients out there talking to the server
> get more chattiness out of the server, the problems described appear
> to be excessive chattiness on the linux nis+ client side.
> 
> Specifically, I have noticed that each nis+ transaction a linux nis+
> client wishes to make, causes it to send ~4 NIS+ RPC NULL PROC calls
> to _each_ NIS+ server. This is for each nis+ transaction. There are
> no similar behaviors seen for solaris clients.
> 
> Looking over the NIS+ client code in glibc, it appears my eyes were
> not deceiving me, there is a section of code designed to send out
> N number of NIS+ RPC NULL PROC calls within a "for" loop.
> 
> 
> On Thu, Oct 31, 2002 at 07:01:16AM -0600, Tom Wike wrote:
> > Here are some more comments on persons that have implemented NIS+ on
> > Linux and their experience on 'chattiness' of NIS+ Linux...
> > 
> > -Tom
> > 
> > On Wed, Oct 30, 2002 at 05:32:13PM -0700, Scott Croft wrote:
> > > I had responded to Tom directly with a question as to whether they were
> > > running compat mode on the NIS+ server. His response back was yes.
> > > Running in compat mode tends to put a lot more traffic over the network.
> > > As you stated, running nscd also helps, but there is a drawback to that
> > > as when you have issues with authentication and possible cache data
> > > corruption.
> > > 
> > > We have A LOT of clients running on our network, AIX, Solaris and Linux
> > > with very little chatter except when needed. I ran snoop on the master
> > > for over an hour and specifically tracked a linux NIS+ system that is
> > > hit very heavily by end users. I had little to no traffic on the master.
> > > On the replica that I know it would hit, there was increased traffic
> > > compared to the master, but no more than the Solaris hosts. We are
> > > running Rh 7.2 with the latest patches and not in compat mode, nor is
> > > compat set on the RH system as some of the documentation suggests.
> > > 
> > > Scott
> > > 
> > > On Wed, 2002-10-30 at 16:52, Robert Edwards wrote:
> > > > 
> > > > I think that, on the whole, the Linux nisplus code is a little "chattier" than 
> > > > the Solaris implementation.
> > > > 
> > > > I have just performed a couple of simple tests (using tcpdump) on some Linux 
> > > > nisplus clients, both with the Name Server Cache Daemon (nscd) running and 
> > > > without. nscd seems to make a little bit of an improvement for, eg., "ls -l" 
> > > > where the UID/GID needs to be looked up from the nisplus maps.
> > > > 
> > > > Just logging into a RH8.0 client as a normal user with home directory mounted 
> > > > from another NFS server (not the NIS+ server) and with nscd running, resulted 
> > > > in 659 packets to and from the NIS+ server. Subsequenty doing an "ls -l" 
> > > > resulted in only another 32 packets.
> > > > 
> > > > Note that our NIS+ server is only a NIS+ server - it doesn't really do 
> > > > anything else (it is a Sun Netra X1 running Solaris 8) - so although there is 
> > > > a relatively large number of NIS+ packets flowing to and from it - this 
> > > > doesn't seems to affect the performance of our network at all.
> > > > 
> > > > We have about 43 Linux and 2 Solaris nisplus clients in our network.
> > > > 
> > > > Cheers,
> > > > 
> > > > Bob Edwards.
> > > > 
> > > > On Thu, 31 Oct 2002 02:52 am, Tom Wike wrote:
> > > > > HI,
> > > > > I could not find any response to this email, we are seeing the same
> > > > > thing and we only have 15 (out of over 200) RedHat 7.2 boxes running as
> > > > > NIS+ clients, this additional traffic is a real concern for us - can
> > > > > anyone comment or tell me your experiences with many Linux NIS+ clients
> > > > > using Solaris NIS+ masters? Thanks!
> > > > >
> > > > > -Tom
> > > > >
> > > > > Marc Wrubleski wrote:
> > > > > >Hi all, I am new to this list, and I couldn't find what I was looking
> > > > > >for in the archives, so please forgive me if this has already been dealt
> > > > > >with somewhere.
> > > > > >
> > > > > >For background, I have configured some Redhat 7.3 (kernels 2.4.18-3 and
> > > > > >also 2.4.18-5) systems as NIS+ clients using nis-utils-1.4.1
> > > > > >
> > > > > >My question is, what would be considered a reasonable amount of traffic
> > > > > >for an idle system with someone logged in using NIS+?
> > > > > >
> > > > > >It seems that our Linux clients talk on the network to the NIS Server up
> > > > > >to 10 times as much as a Solaris client. In a couple of minutes I had
> > > > > >over 1000 packets related to tcp port 32772, NIS+, and portmap on my
> > > > > >system where the only activity is me writing this email.
> > > > > >
> > > > > >Is this normal? If not, what can I do to resolve this issue?
> > > > > >
> > > > > >Thanks in advance.
> > > > > >
> > > > > >Marc Wrubleski
> > > > > >Department of Mathematics and Statistics
> > > > > >University of Calgary

-- 
Tom Wike                                        Email: t-wike at ti.com
ITS Design Systems Support                      Voice: 214-480-2272
Texas Instruments Incorporated                  Cell:  214-212-4497
12500 TI Boulevard, MS 8714                     Pager: 972-648-2012
Dallas, TX 75243-0592                           Fax:   214-480-7676
--------------------------------------------------------------------