[OpenAFS-devel] 50 second fetch-data?

[Harald Barth]

> How many IP addresses has this client used to speak with the server(s)
> in question?

Many. My laptop is travelling often between networks. The server in
question has two interfaces to choose from. Looks to me it chooses
the right one as both that interface and my laptop are in 130.237/16.
The other interface is in 194.132.192/22 but has full routing too,
so it should not matter what interface is chosen as "way out".

> Are the tcpdumps taken from the server side?  If so, are there any
> WhoAreYou queries being sent between the start of the delay and the
> finish?

I have not got any sever side tcpdumps for you :-( I'll try to get
some next time. Now I installed a new server which I can debug better
but then of course the problem vanished for now (see below).

> If so, 1.3.87 does not contain the fix for the UUID to IP address
> mapping problems in which an IP address gets reused by a second or
> third client and the file server merges all three clients into one.
> This was fixed in 1.4.0-rc1.

You probably mean stuff like this:

Wed Oct  5 17:31:21 2005 FindClient: client 8320a78(6d5cb8f8) already had conn a7071568 (host 3fdded82), stolen by client 8320a78(6d5cb8f8)

I have only ONE such log line and not for the time frame in question.
3fdded82 is my laptop 130.237.221.63 when at work. But I have no such
message for any of its other IPs which would be *eded82 (130.237.237.*)
- my laptop when at home.

I moved my H.haba.mail volume to another server which allows me to gdb
and stop the fileserver without been lynched but of course the
problems dissapeared when I did that. Probably I need to use up some
kind of resource in the fileserver/rx first. I don't know how without
letting loose real users. I know I have many connections from many
clients. But a lot of free threads and no CPU or I/O load to speek of.
Feel free to run rxdebug against houting.pdc.kth.se if you think you
see something that I don't. Any tips how to collect statistics?

Harald.