Wed, 14 Nov 1990 17:34:56 +0100
|
The VERS9012 update to BITEARN NODES contains duplicate links for certain
nodes (all of them from BITNET), which could have resulted in critical
operational problems for LISTSERV if we had not been very lucky. My
production server (1.6f) detected an odd number of links in the network
and refused to continue, under the assumption that this meant you have a
node "hanging off the air" or the like. After writing verification execs
and spending 1h trying to understand what was wrong, I discovered that
what we have is a number of duplicate links, as in ':adjnodes1.DUPR (...)
:adjnodes2.DUPR (...)'.
The net effect on release 1.6e is that no error is detected and the
server appears to work properly BUT:
1. Any LINKSWT FILE override for the weight of a duplicated link will be
ignored if the new weight is higher than the normal one (which is the
case most of the time); this is because only the first occurence of
the link will be altered. Luckily, none of the affected links had an
override in LINKSWT FILE; I will let you imagine what would have
happened if transcontinental links had failed to have been weighed
properly and EARN traffic had been routed through the US, or vice
versa.
2. The server will suffer from a degradation in performance due to the
larger amount of possible paths to explore (all leading to the same
result).
3. I cannot guarantee that there will be no other problem, the code was
simply not designed to handle multiple definitions of the same link.
LISTSERV@CEARN crashed with no message whatsoever in the console log
during the tables generation procedure. It subsequently started up
successfully after having been rebooted, but took an inordinate amount of
time (about 10 times longer than normally) to come up. Whether or not
this means that the errors in VERS9012 cause 1.5o to crash is uncertain.
In any case, ALL LISTSERV sites may want to monitor their servers
carefully until the problem is corrected. Here is a list of the affected
links:
-------------------------------------------------------------------------
3231 nodes in file.
Duplicate link: BYUVAX1 -> BYUVM
Duplicate link: CUCSC -> CUHKVM
Duplicate link: DUVM -> VILLVM
Duplicate link: DUVM -> SWATPRM
Duplicate link: DUVM -> DUPR
Duplicate link: DUVM -> DUPHY1
Duplicate link: JNETC -> JNETB
Duplicate link: NCSUMVS -> NCSUVM
Duplicate link: NCSUVAX -> NCSUIE
Duplicate link: NCSUVAX -> NCSUECE
Duplicate link: NCSUVAX -> NCSUCHE
Duplicate link: NCSUVAX -> NCSUCE
Duplicate link: NER -> FSUAVM
Duplicate link: PSUVM -> BKNLVMS
Duplicate link: UICVM2 -> UIUCVMD
Duplicate link: UNTVAX -> UNTVM1
Duplicate link: UNTVAX -> NTSUVAXA
Duplicate link: UTARL -> TAMNET
Duplicate link: UTHOU -> UTMDAPH2
Duplicate link: UTHOU -> UTMDAH
Duplicate link: UTORDOP -> UTORUGW
Duplicate link: VCURUBY -> URVAX
Duplicate link: WELLCO -> WHEATNMA
Duplicate link: WISCPSLC -> WISCPSLB
Duplicate link: WISCPSLC -> WISCPSLA
Total: 25 duplicate links.
-------------------------------------------------------------------------
Two months in a row, now I'm getting curious... What will be UPDATE's
Christmas present to the network? I think I'd better not be around when
VERS9014 comes out.
Eric
PS: I have moved the LSTSRV-M list to SEARN so that its subscribers can
hope to receive this warning before the end of the month.
|
|
|