ASE Cluster Mail Server IMAP+NFS problem

From: Kevin Worvill <K.Worvill_at_uea.ac.uk>
Date: Thu, 21 Oct 1999 12:28:28 +0100

ASE Cluster as Mail server - IMAP + NFS problem
-----------------------------------------------

We have:

Digital UNIX V4.0F (Rev. 1229)
TruCluster Software V1.6 (Rev. 227)

A two node ASE cluster ( Alphaservers 800 5/500 with 2Gb )
with /var/spool/mail on a shared RA 3000 (raid 5) as an ADVFS filesystem.


After running for about an hour the system hangs & a crash dump shows:

message buffer full of:
clntkudp_create: socket bind problem, 48pmap-kgetport: can't create RPC handler.

Network mail connectons are:

# grep imap crash-data.0 | wc -l
       490
# grep imsp crash-data.0 | wc -l
       100
# grep pop crash-data.0 | wc -l
        27
# grep exim crash-data.0 | wc -l
        10

imap is the ESYS/ExecMail (Simeon) version IMAP2bis Service 7.8(89isa)

ESMTP Exim 3.02 #1

POP3 3.3(18) from UWash

Simeon IMSP 1.3


Compaq suggest that the IMAP connections are taking all the ports available
to NFS ( which appears to be a hard wired kernel limit ) and this is then
hanging NFS.

Has anyone else seen similar problems - have a solution or suggestions ?

To workaround we are currently running IMAP on the host having the spool volume directly
mounted & avoiding the NFS mounted path with a link - but this breaks the
ASE rules for an NFS service & makes failover difficult. Our average IMAP
load is around 1000 connections.


-----------------------------------------------------------
Kevin Worvill phone : +44-1603-592392
IT and Computing Services fax : +44-1603-593467
University of East Anglia email : K.Worvill_at_uea.ac.uk
Norwich, NR4 7TJ, UK
-----------------------------------------------------------
Received on Thu Oct 21 1999 - 11:30:43 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:39 NZDT