Eric Thomas <ERIC@FRECP11>
Fri, 1 May 1987 21:54 SET
|
I forgot to mention two other "design deadlines". These have nothing to do
with human intervention, it's just a technical goal I would like to reach.
- Delay between the time a DISTRIBUTE job is posted and the time it has
reached 90% of its recipients, with average network load (around 500 on
BITNIC -> EARNET and 100-200 on CEARN <-> DEARN): this depends on the size
of the file of course, but I would like this to be one fifth of the time
SENDFILE would take on the same file for 30 recipients. However, the
absolute deadline should be the time SENDFILE would take. This is why I
think it is important to have the backbone servers restarted in case of
failure. And this is why I cannot accept nodes which place the server
offline during peak hours, even if it is only 3 hours a day. Otherwise users
will get nervous and will revert to SENDFILE :-(
- Time required to PUT a (relatively small) file on 95% of the backbone
servers, under the same conditions as above: 24 hours. List owners should
not have to wait for more than 24 hours to have an update to their list
propagated to all the peers. It should be possible to remove a server from
the backbone in about 1 day if it is scheduled to be down for some period of
time.
A NETSERV PUT takes about 2 hours to reach all non-down servers, except
CANADA01 and TCSVM. It takes about 8 additional hours to reach those nodes.
There are much more LISTSERVs than NETSERVs, so we cannot expect the same
speed, but 24 hours would seem to be quite feasible.
The last time I distributed a small file (PROBLEMS MEMO), it had reached 75%
of its recipients in 12 minutes under very good file queue conditions. And it
took about 3 hours to reach all the nodes that were not down for the whole
weekend. But then it took several days to reach those few nodes. That's why
I'm always speaking of 90% and suchlike.
Eric
|
|
|