RSYNC Hangs on large file counts and xinetd

[Bachmeier, Marc]

Got a weird one, wondering if anyone has any ideas

Running an RSYNC 2.6.3pre1 server on Solaris 9 (Single USIII CPU, 1 GB
RAM, SAN Disk), client is 2.6.3pre1 on Solaris 8 (Dual USII CPUs, 2 GB
RAM, SAN Disk). In between the two is a Solaris 9 (Single USIIi CPU 1 GB
RAM, Local Disk) system using xinetd to forward traffic from the client
to the server on port 873.

The client is configured to push files by directory to the server via a
cron job. The directories that are having the problem contain network
element performance data files. These directories can have up to a 1.4
million files in them, but the files are very small in size. 

The current setup that is having the issues is as follows:

Client -- TCP 873 --> xinetd server -- TCP 873 --> Server

The symptoms are that the file list will transfer, the checksum
comparison completes, and client begins to send files. However, the
client will only be able to send a few files (anywhere from 2 - 200),
and then the rsync completely hangs - CPU drops to .01% on both sides,
but the connection is still open.

This configuration works fine when the xinetd server is pulled out of
the picture. Unfortunately, I can't remove it permanently - it's
required to get data from one network to another.
 
Any ideas?

Marc