jsdy@hadron.UUCP (Joseph S. D. Yao) (01/10/89)
I'm hoping that this will sound familiar to someone, who will then
send me mail (since I don't get this far in my .newsrc often).
A problem was reported to and observed by me, in sending certain mail
messages from VAXen running Ultrix 2.2 to a VAX running System V
Release 2 Version 2 (modified by me) with Uniq's Passage software
(again modified by me in places, to make it work at all). After a
number of observations, it became obvious that only "large" (not yet
quantified) messages were failing, and the SMTP program on the Uniq
side was getting an error message, "URESET - network connection
aborted and reset". Similarly large messages worked fine among Ultrix
2.2 machines.
The Ultrix 2.2 machines are running a - possibly early - version of
BSD 4.3 network code, while the Uniq code appears to be slightly
modified BSD 4.1.
Instrumentation of the Uniq side shows that the TCP level is receiving
several packets perfectly reasonably, then all of a sudden getting a
packet whose sequence number is 1024 more than expected! At this
point, of course, it stops and waits for the next one ... which never
arrives. This 1024 over is not a bit set - it is whatever-odd-number
plus 1024 (i.e., there is bit carry and so forth). Since attempts to
re-receive the "proper" packet fail, it then times out and resets the
connection.
Any takers? Mucho obligato if so. Thanks to all...
Joe Yao jsdy@hadron.COM (not yet domainised ???)
hadron!jsdy@{uunet.UU.NET,dtix.ARPA,decuac.DEC.COM}
arinc,att,avatar,blkcat,cos,decuac,dtix,\
ecogong,empire,gong,grebyn,inco,insight, \!hadron!jsdy
kcwc,lepton,netex,netxcom,phw5,rlgvax, /
seismo,sms,smsdpg,sundc,uunet /mike@BRL.MIL (Mike Muuss) (01/11/89)
Two suggestions: 1) check for trailers handling. (use "ifconfig xx0 -trailers") 2) check the offered TCP MSS -vs- what is getting sent. A Sun running "TCPDUMP", or an ethernet analyzer, will be helpful. Best, -Mike
jsdy@hadron.UUCP (Joseph S. D. Yao) (01/11/89)
In article <828@hadron.UUCP> I (Joseph S. D. Yao) write: >... >Instrumentation of the Uniq side shows that the TCP level is receiving >several packets perfectly reasonably, then all of a sudden getting a >packet whose sequence number is 1024 more than expected! ... Many thanks to the several people who immediately suggested an answer that seems to exactly fit the problem: Ed Frankenberry <ezf@BBN.COM> Doug Nelson <08071TCP%MSU.BITNET@CUNYVM.CUNY.EDU> cpw%sneezy@LANL.GOV (C. Philip Wood) map@gaak.LCS.MIT.EDU (Michael A. Patton) "Mark D. Eggers (219) 239-7258" <CF4A8X%IRISHMVS.BITNET@UICVM.UIC.EDU> Mike Muuss <mike@BRL.MIL> Apparently, as I'd suspected, Ultrix 2.2 TCP/IP is 4.3BSD-alpha. It offers "trailers" by default. If a large packet is sent, 1024 bytes are sent in a "trailer", which the Uniq (4.1BSD) TCP or IP (I think the latter) doesn't recognize ... hence my symptom of suddenly finding sequence numbers boosted by 1024. The universally proffered solution was to put "-trailers" as an argument to 'ifconfig' in the system initialisation file "/etc/rc.local". It was suggested that the Ultrix implementation of trailers, being 4.3-alpha, might be incomplete; or that Uniq TCP/IP, being older, probably doesn't understand trailers. This is not surprising. I'd already changed the net broadcast address to all-zeroes, since the Uniq implementation of all-ones is incomplete. I'll try this tomorrow. If you don't hear from me again on this, this is the final answer. Mike Muuss, as always, thinks up more possibilities. I don't under- stand his second suggestion, so I'll include it verbatim and let him elucidate, if he wants. > 2) check the offered TCP MSS -vs- what is getting sent. He also adds: > A Sun running "TCPDUMP", or an ethernet analyzer, will be helpful. Joe Yao jsdy@hadron.COM (not yet domainised) hadron!jsdy@{uunet.UU.NET,dtix.ARPA,decuac.DEC.COM} arinc,att,avatar,blkcat,cos,decuac,dtix,\ ecogong,empire,gong,grebyn,inco,insight, \!hadron!jsdy kcwc,lepton,netex,netxcom,phw5,rlgvax, / seismo,sms,smsdpg,sundc,uunet /