really odd server timeout problem

[danix]

I doubt this is caused by Zimbra, but here goes. I'm at my wit's end with this one.

We're been running a zimbra server for a few weeks.
Recently we noticed that from our office network (and only from this network) we were seeing random timeouts, like "server is not available".
I put up a hobbit monitor in the office, as well as one outside, and can draw the following conclusions:
1) http, https, imaps, and even ssh become unavailable from our office monitor. Timeouts are confirmed by manual telnet to the above ports.
2) At no other time do we see any outage of these ports from any other network other than our office network.
3) Traffic from the zimbra back to the office network works during the outage.
4) We've checked the office systems for virus, IP conflicts, arp conflicts.
5) We've replaced the office firewall as a test, it didn't help.
6) The problem can be reproduced on any PC inside the office LAN, macs, linux, windows, etc. but not consistently.
7) There is no obvious correlation with time of day, though it seems to happen less often at night and on the weekend. We are not exceeding the limits of our firewall or internet connection.

Running tcpdump on the zimbra server shows the syn packets coming in from the office network, but no acks go back (when the problem is happening). They just vanish.

The zimbra server is CentOS 4.6 with Zimbra 4.5.10_GA_1575.

Thanks.

[jholder]

Is your router being flooded by NAT requests? I've seen where the router can be flooded to the point it drops packets.

If running cisco, run
#ena
#show ip nat tran *