Subject: | |
From: | |
Reply To: | |
Date: | Mon, 30 Sep 1996 22:17:38 -0700 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
gary flynn wrote:
> Our three telnet express boxes are intermitently dropping connections.
> The problem appears to be related to router CPU and/or traffic
> load. From what I've seen with a protocol analyzer, the box locks up
> and eventually sends a RST to the connecting PC which is usually running
> Reflection for Windows. Other connections are not having problems although
> things slow down somewhat. I haven't attempted to decode the proprietary
> stuff between the Telnet Express and the 3000. Does anyone know how robust
> the TCP implementation is on these boxes?
The 3000 is not very forgiving on longer timeouts, such as a congested
network or distant link; but I have no empirical evidence to support that.
If you are using Reflection's stack, it is not that robust on certain upper
layer error handling. I did isolate one fault in WRQs stack up through
the 5.0 version (maybe later; but I reported it and it was patched) where
an ICMP redirect would close a connection (treated like an RST). In your
case of a busy router, an ICMP source quench might cause similar quirks.
While the two (WRQ and 3000) "kick butt" on a local network, I'm not quite
so convinced of their tuning for distant connections with considerable
latency, nor slow links. Just my own opinion and "feel" about it; again,
no empirical evidence other than the documented ICMP redirect quirk (by
the RFC definitions they should be treated as a transient error and the
packet retried, not reset the connection).
Jeff Kell <[log in to unmask]>
|
|
|