brian@ucsd.EDU (Brian Kantor) (05/18/89)
I recently installed sendmail 5.61, and after a week of incredibly bad performance, went back to 5.60 while I figure out what's going wrong. The problem was that our throughput dropped tremendously. We have only a Vax-750, which you might optimistically rate at .75 MIP, and the new sendmail seemed to make the system run quite a bit slower. I suspect that it is 5.61's interaction with the nameserver that causes this. We have it configured with the NO_WILDCARD_MX option set, which causes it to do a C_ANY call on the nameserver instead of a C_CNAME. I believe the theory behind doing this was that by making such a call, the local nameserver would cache all the data for the hostname being looked up, which would save a second reference across the internet for the common case where an MX lookup is followed by an A lookup for the same host. However, this also causes WKS, HINFO, and every bloody thing else known about that host to get transferred too, and we have observed that the memory working set of the bind nameserver with 5.61 was much larger than with 5.60 which didn't do C_ANY. In fact, it got larger than our available physical memory, which means that the poor little 750 was paging for a good percentage of the mail passing through. I have yet to try 5.61 with the C_ANY disabled, as it took us several days to recover from the previous misadventure and I'm not anxious to go through that again. Does anyone have any ideas about this sort of interaction? Or is there simply some bugfix for 5.61 that'll correct this situation? Brian Kantor UCSD Office of Academic Computing Academic Network Operations Group UCSD C-010, La Jolla, CA 92093 USA brian@ucsd.edu ucsd!brian BRIAN@UCSD