west@sdcsla.UUCP (Larry West) (12/21/84)
[Yes, Lauren, please proceed as you see fit.]
I'm curious as to how the satellite-broadcast scheme will take
power-failures and the like into account.
For example, is reception dependent upon the host computer being up?
[I.e., I assume the thing doesn't have its own disk.]
Although currently a site can stand missing many many hours
of netnews reception, I wonder whether this will be true
in the satellite system. The problem that I see is that
the satellite system will become the backbone of netnews
distribution -- so, if a satellite (reception) site is down
for a while, whence will it get the news it missed?
I guess arrangements with the other satellite reception sites
can be made (at the expense of long-distance phone calls),
but... is there A Better Way?
--
--| Larry West, UC San Diego, Institute for Cognitive Science
--| UUCP: {decvax!ucbvax,ihnp4}!sdcsvax!sdcsla!west
--| ARPA: west@NPRDC { NOT: <sdcsla!west@NPRDC> }lauren@vortex.UUCP (Lauren Weinstein) (12/23/84)
While it would be possible to get netnews from a nearby site, there is a better way to deal with longish down periods. Under normal conditions, the gateway computers should be able to hold a week or two of news online (this is an estimate, of course). So long as the host down period doesn't exceed this interval (which should be fairly rare) my plan is as follows: Part of the information being sent by stargate would include (at intervals) a list of sequence numbers in the current transmission sequence, and the earliest sequence number still on disk at stargate. If a site discovered that it was missing articles outside the current sequence but still online, it would send a specially formatted mail message (via UUCP) to stargate. Stargate would then reschedule (as time permits) the retransmission of those missing articles for that host for their pickup. In other words, individual sites could request that stargate temporarily add "old" articles back into the current sequence for their reception. I think this should deal with most of the problems with host down periods. --Lauren--
eric@milo.UUCP (Eric Bergan) (12/24/84)
Given that there are now hundreds (thousands?) of news machines, is it really reasonable to try and send "makeup" news whenever one of them is down? Seems to me that at any given moment, there must be at least a handful of news machines down. Perhaps a better solution is simply to allow a machine that has been down to call a "news repository" via UUCP and get the missing articles. Then the burden is placed on the system that went down, not the satellite distribution. I suspect the result would be that some sites would determine it is not important to get the missing news, while others would be allowed to catch up if they wished. In either case, the software should be set up so that the site administrator is presented a list of groups, and missing articles, so that they can decide not to pick up the 1236 net.flame articles, but they do want the 15 net.sources articles. Perhaps we should somehow design this to be at least partially compatable with the "distributed news" effort? Seems there are some common traits, albiet on different scales. -- eric ...!seismo!umcp-cs!aplvax!milo!eric
lauren@vortex.UUCP (Lauren Weinstein) (12/25/84)
There may not be a serious problem with many machines requesting repeats. For example, during the periods of higher bandwidth on the stargate (presumably during the night) the stargate could routinely send older articles that have been requested by any number of sites during the day -- or even some older articles that weren't requested at all, just on general principles. There are all sorts of possibilities. Of course, sites could also pick up lost articles from other sites if they wanted to, or administrators could decide that there were some articles (or groups) that they didn't want to pick up (if lost due to system down time) at all. The important thing to remember is that it doesn't cost any more to send an article to ALL participating sites than to just one site using the bird. It might well be good policy to repeat older articles (those still online) occasionally just to help people fill in any gaps. Much of this will need to be determined experimentally of course. ---- By the way, the stargate screeners mailing list is up. It is seismo!stargate-people. It is for people willing to screen netnews groups for the project, especially people whose sites are willing to call stargate (in Atlanta) directly. --Lauren--