Retieval of files from Samba is erratic : solved

[Alex at Avantel Systems]

At 02:59 PM 3/22/1999 -0500, David Collier-Brown wrote:

>Could you send this to the mailing-list and/or the newsgroup?
>It's an interesting thing to have in the searchable discussions...

Original Problem: After a minimal setup of smb.conf, browsing worked fine
and files could be saved on the samba shares but when a file was retrieved,
some would retrieve OK, others would not. A great variety of error messages
were encountered. Several complained about wrong path names, and some
indicated that the network resource was no longer available. In several
cases win95 froze and required a re-boot. File retrieval was tried in a DOS
window, with explorer and by retrieving files into applications.

Test procedure (over a 2 month period)
1)Passed ALL tests in the troubleshooting guide included with the samba
distribution, in Blair's Samba reference, and in yet-to-be-published test
procedures.
2) Used tcpdump (smb version) to trace traffic. Copy sent to Samba team -
no abnormal behaviour identified. 
3) Since abnormal win95 client behaviour was suspected, compared versions
of the many variations of W95 (using listqfe and win95ver) to systems known
to work - all up to date. (My appologies to Steve Arnold, I lost the url
for these test files that you suggested)
4) Tried with 2 other clients (one on each of two subnets) one passed and
one failed.  (This really led me astray - I still do not know why one
passed! Perhaps not tested with enough different files?)
5) Since the client on the other subnet passed, I exchanged one nic for the
other in the samba host (same type of nic) with NO change in behaviour. The
clients continued to behave exactly in the same way as before the switch.
6) Tried ftp; ftp TO the samba server always worked. From the samba server,
ftp of small files worked OK but large files failed. (windows locked up in
almost all cases).  Ftp to/from other hosts worked fine.
7) Tried pings with larger (maximum) packet size - failed test (sporadic
timeouts/packet loss). Large packet pings to other hosts worked fine.
8) On large ftp transfers from the Samba host, tcpdump showed repetition of
many of the initial packets and then continuous repetition of the same
packet without response from the Win95 ftp client (usually frozen).

Solution: Concluded that the NIC was having trouble with transmitting
(formatting/buffering/other?) full packets, replaced it, and Samba works!
NIC may be less than 100% ne2000 compatible.

Post Mortem: Could have caught this sooner in steps 1 (with larger ping
packet sizes) and 2 (it showed the same behaviour as step 8 above).  Step 5
has me baffled but would not have helped even if the results had been the
same for both clients.  Perhaps I was too willing to blame Win95. However,
win95 should certainly be able to handle packet corruption more gracefully.  

Alex
Avantel Systems
Ottawa, Ontario, Canada