JRE VM dying under random network events

From: diggum (durin_at_duringleaves.com)
Date: 12/31/03


Date: 31 Dec 2003 10:47:02 -0800

Hi,

We've been troubleshooting a connection problem with a client of ours
for several weeks. Basically, a combination of network stuff appears
to be causing the VM to freeze, or at least just kinda wander off for
awhile.

Our client software pings our server roughly every 30 seconds. This
customer is connecting using HTTPS Tunneling. They're running through
their proxy server and hitting their ISP, routing to us via MCI's
backbone.

There's an MCI router in Salt Lake that consistently shows around 10%
packet loss average. At random points throughout the day, the
connection is dying and unable to restore within our 3 minute timeout
period. On most systems and scenarios, the client software recognizes
this and notifies the user allowing them to reconnect.

Under THIS particular scenario, however, quite often the VM simply
stops - no more attempts to ping, no acknowledgement that the
connection is dead or that no response has been received, and nothing
else happens until the applet is closed. it can be restarted
immediately and the connection will be fine and happy. ALL logged in
users are disconnected at the same time, and some see it right away,
others do not.

I HAVE discovered that most of the users who appear to have the dead
VM have 1.3.1_03 AND 1.4.2_01 installed on their system, though only
1.4.2_01 is running at the time.

If anyone else has seen something like this, please let me know. We're
attempting to recreate it inhouse by running through a proxy and
simulating the approximate packet loss they experience (since MCI has
been unwilling to admit there's a problem with their server regardless
of the ping plots we give them) but they have a non-mainstream proxy
server.

Thanks
Durin