What kind of performance difference are you seeing?
Are you seeing evidence of dropped/retransmitted packets, particularly
in the XP trace? By turning on Preferences->Protocols->TCP->Analyze
TCP sequence numbers, this should be fairly easy to spot (you should
see a number of frames marked with "TCP previous segment lost" and "TCP
retransmission" flags, among others).
Take a look at SMB statistics by going to Statistics->Service Response
Time->SMB (The SMB stats may be in a slightly different place if you're
running an older version, in which case you might want to consider
upgrading). This gives you a series of useful info - the number and
type of SMB calls made, as well as the amount of time it appears that
the server took to respond to each call.
Is there a significant difference in the workload (types and amount of
calls) generated by the Win98 client versus the WinXP client? Does the
server seem to be significantly slower (taking a longer time to
respond, i.e. higher Server Response Time) in one test than the other?
Keep in mind that, unfiltered, this data is for ALL SMB conversations,
so if the system you're capturing on was talking to more than one CIFS
system at once, then the stats here might be skewed (in which case,
just filter on the IP(s) of the system(s) you're interested in)
As you look through the trace do you see any significant delays (long
delta times) between packets? If so, try to figure out why these are
happening (could be because of packet loss)?
SMB calls generated by Win98 can differ significantly from WinNT/2K/XP
systems running the same app - there are particularly significant
differences when running 16-bit apps (which I'm assuming your "DOS
application" is). Usually, though, XP clients tend to be faster...
Are reads and writes (copies from and to the SMB server) also
significantly faster or slower?
Ian
On Jul 20, 2004, at 2:53 PM, Andy Bender wrote:
Thanx for the tip, Simon. I checked the packet capture set for the
XP<->98SE
session, and didn't see any SMB FLUSH PDUs. In order to simplify the
analysis, I've isolated a single shared table read operation that
simply
opens the table, and reads a few records. No sign of any flush
requests,
which is logical, given that caches writes wouldn't be required.
Interestingly, to complete the same operation in a 98SE<->98SE session,
Ethereal captured 27% fewer frames. Seems that somehow 98SE is more
efficient in this scenario. I wonder why.
APJ.
----- Original Message -----
From: "Simon Hailstone" <Simon.Hailstone@xxxxxxxxxxxxxxxxxxx>
To: "Ethereal user support" <ethereal-users@xxxxxxxxxxxx>
Sent: July 20, 2004 2:34
Subject: RE: [Ethereal-users] Why is WinXPPro SMB slower than Win98SE
SMB?
Howdy, y'all; I'm trying to determine why a 166 MHz Win98SE
machine can, from a DOS box, open a file on a 98SE server via
SMB in less time than can a 1.3 GHz XP client running the same
DOS app.
Hi Andy,
In the trace of XP<->98SE Server, do you see SMB Flush requests
being sent from the XP host to the 98SE Server? I often see them
when transferring data between 98 and Win 2k, with one FLUSH
occuring every 30-40kB or so. They have a very detrimental
effect on performance.
Follow this reference for more details, if you discover that you
are in fact seeing this problem.
http://manubatbat.free.fr/doc/smb/4.2.8.htm
Unfortunately, I cannot offer a solution to the problem, merely
a potential explanation.
Best Regards,
Simon Hailstone
Orthogon Systems
_______________________________________________
Ethereal-users mailing list
Ethereal-users@xxxxxxxxxxxx
http://www.ethereal.com/mailman/listinfo/ethereal-users
_______________________________________________
Ethereal-users mailing list
Ethereal-users@xxxxxxxxxxxx
http://www.ethereal.com/mailman/listinfo/ethereal-users