Did you eliminate broken PMTU discovery as a source of the problem. There is a Pacbell DSL box (Redback) that is my immediate next hop and it randomly decides to drop the MTU down to 1400 bytes or so every now and then. Since you mentioned DSL customers, you may want to check by putting in a sniffer that you are not getting PMTU discovery process all messed up. On Friday, September 7, 2001, at 07:52 AM, Nanog wrote:
Ok guys,
Interesting problem that seems to have started on Friday.
Here's the deal, it seems that certain (not all) web pages on various sites we host are not loading for certain customers. It started on Friday with visitors on Verizon DSL waiting forever for pages to load.
Yesterday and today PacBell DSL and some Netcom customers started complaining.
This only seems to occur on HTTP traffic. If we have the visitors try HTTPS the pages load fine (with the normal encryption slowdown).
We've looked at Layers 1-4, and can't see any problems, ping looks great, interfaces and cpus on routers, servers and switches look fine.
It's almost like it's a transparent cache "bug". Anyone know if Inktomi, or any major cache vendors rolled out any new code this week?
For that matter does anyone know what caches Verizon and SBC use for starters?
Other variables: it only seems to occur on IIS based systems (I know I know, no flame wars or suggestions for replacements). Although it's not happening on all of our IIS servers.
Any clues are greatly appreciated.
Steve