sverre@fesk.UUCP (Sverre Froyen) (12/12/87)
After posting the message about the broken bitnet gateway at super, I got a number of replies asking why I did not simply use the gateway at psuvax1. Why indeed. The reason was, of course, that pathalias generated the path from the recently posted uucp maps. Still it seemed strange that it should prefer a long path like "boulder!hao!husc6!harvard!gymble!super!%s" rather than the shorter path, "boulder!hao!rutgers!psuvax1!%s". A little studying of the map files revealed the reason: Super gives the cost of the bitnet link as DIRECT whereas psuvax does not give a cost. No cost is equivalent to 4000, rather higher than the DIRECT cost of 200, and certainly more than enough to offset the (slightly) higher cost of reaching super compared to psuvax1. When I then learned that all super does, is forward the mail to psuvax1, it made the automatic routing somewhat meaningless. In conclusion: (1) Don't list a gateway unless you run a true, honest to god gateway. And (2), if you list a gate- way, don't associate a cost with the gateway. (Perhaps pathalias should be modified to ignore gateway costs). Sverre -- Sverre Froyen UUCP: boulder!fesk!sverre, sunpeaks!seri!fesk!sverre ARPA: froyen@nmfecc.arpa BITNET: froyen@csugold.bitnet
nsadmin@egg-id.UUCP (Linn Hower) (12/13/87)
> A little > studying of the map files revealed the reason: Super gives > the cost of the bitnet link as DIRECT whereas psuvax does > not give a cost. No cost is equivalent to 4000, rather > higher than the DIRECT cost of 200, and certainly more than > In conclusion: (1) Don't list a gateway unless you run a > true, honest to god gateway. And (2), if you list a gate- > way, don't associate a cost with the gateway. (Perhaps > pathalias should be modified to ignore gateway costs). > > Sverre > -- > Sverre Froyen I also had bitnet mail bounce and tracked down this same problem. However I don't agree with the above conclusion. If I am providing a gateway service on a single computation node, its `cost' is very low, approaching zero depending on my cpu horsepower. Why don't we use pathalias's input for what its designed for? Why have a special case for gateways? I feel the input from the d.Top file should reflect the true cost of gatewaying. ( I am not arguing the error in super's entry. Its pretty screwed up.) -- Linn -- Linn Hower usenet@INEL.GOV Phone: 208-526-9353 usenet%INEL.GOV@uiucuxc.ARPA
matt@ncr-sd.SanDiego.NCR.COM (Matt Costello) (12/15/87)
In article <567@egg-id.UUCP> nsadmin@egg-id.UUCP (Linn Hower) writes: > I also had bitnet mail bounce and tracked down this same problem. >However I don't agree with the above conclusion. If I am providing >a gateway service on a single computation node, its `cost' is very >low, approaching zero depending on my cpu horsepower. Why don't >we use pathalias's input for what its designed for? Why have a special >case for gateways? I feel the input from the d.Top file should reflect >the true cost of gatewaying. The definition of gateways should almost always use the default pathalias cost of 4000. Even though a gateway is defined as a link (to pathalias) its real cost is always 0 because that machine IS the gateway. While the published maps stick to the default cost for domain gateways I'll be able to locally override these costs for specific cases. Right now, for example, relay.cs.net is listed within NCR as a gateway to bitnet with a cost of 95; This was done to override super's incorrect map entry. Our gateway to .MIL is a nearby machine actually on the MILNET. Please people, don't put everything out into the USENET map files. These files are supposed to be getting smaller as domain mailers become more common. I've got systems that are sometimes unable to use the full USENET database because the user process size limit is only 1 MB. -- Matt Costello <matt.costello@SanDiego.NCR.COM> +1 619 485 2926 <matt.costello%SanDiego.NCR.COM@Relay.CS.NET> {sdcsvax,cbosgd,pyramid,nosc.ARPA}!ncr-sd!matt