Hello out there in Ethereal Land,
I'm not sure how much my question belongs here vs. a more general
networking newsgroup, but I'm a bit of a network newbie, so if it
belongs elsewhere any advice on where would be appreciated and apologies
for cluttering the list.
I'm using ethereal to debug a performance problem on my LAN. I have a
Win2003 Server (which is also being used as a Domain Controller, Active
Directory, etc. It's the all purpose small office server) and 2 Windows
XP clients. The server hosts a file share with files for my
application. Each PC runs a thick client for the app (so the server is
really nothing but a file share as far as this app is concerned).
When doing a search in my app, I've found that the first PC (call it
PC#1) to attach to the share gets relatively good performance (2 second
response). The second PC (PC#2) then slows way down (7 second
response). Note that the results are the same regardless of which
actual PC I use as PC#1 and PC#2 - it's the order and not something
about the actual PC that changes the performance.
My ethereal trace showed me that the extra slowdown on PC#2 is caused by
lots and lots of extra SMB Packets. (PC#1 has around 20% SMB packets
and the rest general TCP/IP, whereas PC#2 has around 80% SMB packets.
Around 4MB of extra data passing back and forth.). The surplus seems to
all (or almost all) be of the form
Read Andx Request
Read Andx Response
At one point there are around 22,000 consecutive such transactions with
almost nothing else.
Interestingly enough, when I make a duplicate copy of my application
data and host it on my laptop (XP) rather than the Win2003 Server, the
problem goes away. That is, PC#1 and PC#2 can simultaneously connect to
my laptop and get almost identical response times with nearly identical
ethereal traces. I have swapped cabling & network ports between my
Laptop and the Server to help rule that out as a cause. So there is
something about that Win2003 Server (and the way it interacts with the
world) that is at the root of the problem.
NetBios is enabled, although I disabled it on a PC (but not on the
server) to force the traffic outside of Netbios and port 139. That did
not substantively affect the results of the experiment.
Any ideas of what might cause this or steps you might take to try to
narrow it down (or as I mentioned before another place I might want to
post) would be greatly appreciated. I'm a bit at a loss.
Thanks!
--Tom
