httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chuck Murcko <ch...@n2k.com>
Subject Re: Finish 1.2!
Date Fri, 10 Jan 1997 08:53:12 GMT
Cliff Skolnick liltingly intones:
> 
> On Fri, 10 Jan 1997, Chuck Murcko wrote:
> 
> > It's the client TCP stack never sending a final FIN. The people moaning are
> > ones who can't set a timeout for FIN_WAIT_2 in their kernels (BSDI users who
> > don't have kernel source, and SunOS so far, but RobH's biting the bullet and
> > dealing with it). I just took yet another look at our internal Apache proxy,
> > and every single FIN_WAIT_2 connect is from a WIN95 box. Not Solaris, not
> > IRIX, not Linux, not FreeBSD, not BSDI, not WinNT, and not Win3.1 running
> > Trumpet Winsock. Yes, we run all those here.
> 
> OK if it is win95, I fear there may be some data timing issue here that 
> tweaks a buggy 95 stack.  Add choice #3, and this one would be really 
> hard to trace down.  This is a bug for someone with a good pair of hip 
> boots...
> 
> Chuck - one question - 
>   Are those permanently connected win95 boxes or are they dial-up users?
> 
They're permanently connected. It occurred to me that while it's possible
they're left behind on machine crashes, there are just too many of 'em
for that.

That's not what I'm seeing in the front end Web server logs, although the
largest precentage of FIN_WAIT_2 connects are Win95, there's a smattering of
nearly everything in there. They are mostly Mozilla, all 3.x. Both
regular and Gold. Also a fair number of MSIE 3.x on Win95. It looks like
the deadliest combo is Mozilla 3.x/Win95, followed by MSIE 3.x/Win95.

Mmmm. Keepalive timeouts?

Mozilla/3.x and Win95 is all of our internal FIN_WAIT_2 states, and about
50% of our external ones by itself. Another 15-20% are MSIE 3.x/Win95.
OTOH, that's a fairly good approximation of what the user setup of choice
is these days.

BTW, we also have Macs in house, but none of them are in the proxy's
FIN_WAIT_2 state list.

chuck
Chuck Murcko	N2K Inc.	Wayne PA	chuck@telebase.com
And now, on a lighter note:
There are no data that cannot be plotted on a straight line if the axis
are chosen correctly.

Mime
View raw message