enger%gburg.DECnet@BLUTO.SCC.COM ("GBURG::ENGER") (08/04/87)
I experienced some difficulty connecting to the NIC on Sunday night. This led me to try to PING off of them (issue ICMP echo request). I did not receive any response from them on either their net-10 or net-26 address. I was able to receive ICMP echo replies from hosts on a number of networks so I wrote it off as a temporary network problem. I PINGed them again on monday, still no response. I called the NOC. They used TELNET to test connectivity, and they got through. The NOC was also able to TELNET to my host. When I told them I couldn't PING off the NIC, they became concerned. Taking their lead, I tried to TELNET to the NIC, and lo I got through as well. Yet, when I closed out the session, and tried PINGing them again, still nothing. I called up the NIC. They took down my complaint, and the next day someone from engineering called me back. He said the NIC has turned off their ICMP echo replies. He said they were getting too many ECHO requests and that it was loading the machine down. ICMP ECHO request/reply has been a usefull debugging tool. I hope this doesn't start a trend. ------
pogran@CCQ.BBN.COM (Ken Pogran) (08/05/87)
Regarding PINGing and the report that the NIC has elected to turn
off ICMP echo replies ("He said they were getting too many ECHO
requests and that it was loading the machine down. ICMP ECHO
request/reply has been a usefull debugging tool. I hope this
doesn't start a trend."):
This issue is only about thirteen or fourteen years old. Perhaps
Mike Padlipsky will supply us with the correct bibliographic
reference; back in the NCP days he wrote an RFC entitled
something like, "... but my NCP costs $500 a day!" complaining
about incessant NCP ECHO requests from TENEX hosts that were
causing the MIT-Multics Network Daemon to wake up constantly to
send ECHO REPLYs. The solution for us at MIT, back then, was
two-fold: 1) Get the TENEXes to stop frequent pinging which was
being done for the sole purpose of keeping track who was really
up on the ARPANET and who was not, and 2) move the Multics
ECHO-processing code into an interrupt handler and out of a
process that needed to be awakend for each echo. Both were
accomplished.
Yes, pings are nice. They provide you with assurance that
someone is really there. As the Internet grows though (and we're
over 250 truly active nets, now), unnecessary or gratuitous pings
are a waste of everyone's cycles -- hosts, gateways, PSNs. And
at a well-known, heavily-used host like the NIC -- imagine the
horror experienced by a system manager who, trying to respond to
user complaints of slow service, does a profile of where his
cycles are going and discovers that a substantial fraction is
going into ICMP ECHO replying! (I haven't talked with the NIC,
so my description here is purely hypothetical.) Here, clearly,
is a way to "buy back" cycles that can be used to improve service
to users.
Is it going to start a trend? Given the ever-increasing number
of nets out there, we might indeed begin to see more "defensive"
moves made by major service hosts who begin to perceive completely
open, full, and friendly participation as a drain on their
resources.
Food for thought.
Ken PogranMills@UDEL.EDU (08/06/87)
Bob, I can't believe this. Nobody (!?) pings the NIC unless they can't connect and wants to find out why not. So the NIC turns off ICMP as a defense measure? My prior experience has been that ICMP worked, but TCP connections took a v e r y long time to complete. It is certainly obvious that the NIC is overloaded; however, if a significant contributory factor is ICMP echoes, I submit turning them off will only result in their replacement by TCP connection attempts. Considering the TOPS-20 initial-connection TCP design, this would be an even worse disaster. Dave
hedrick@TOPAZ.RUTGERS.EDU (Charles Hedrick) (08/08/87)
That's certainly true for us. Since the NIC is not a gateway, the only reason we would ever ping it is if we didn't get response from the root name server and wanted to help diagnosi
phil@BRL.ARPA (Phil Dykstra) (08/09/87)
Last time I looked at traffic from the NIC they seemed to be spending most of their time sending ICMP Source Quenches in response to nameserver requests. The nameserver system is probably the major reason for their load. Some statistics gathered a month or two ago showed them fielding an average of four queries a second. - Phil
snorthc@NSWC-OAS.ARPA (08/10/87)
I have seen several replies from users saying they only ping the NIC when they can't make a connection. I was just looking through a manual for TCP for PCs. The example they give for ping is: ping -t sri-nic.arpa where -t is an option to ping repeatedly. Stephen Northcutt (snorthc@nswc-g.arpa)
PADLIPSKY@A.ISI.EDU (Michael Padlipsky) (08/10/87)
Ken [and interested onlookers]-- I am appalled to have to report that I can't find "But My NCP Costs $X00 a Day..." despite the facts that 1) I could have sworn I'd done such a thing and 2) running my finger down all the lines of RFC 1012 should have confirmed it. Either it was done as a Project MAC Memo and didn't get out on the net, or it's hidden away under some other title (RFC 411, "New Multics Network Software Features," is a possibility, but I can't get through to the NIC to check up), or (shudder) perhaps there's an error of omission in RFC 1012.... Of course, the incident is immortalized, albeit in highly capsulized form, in The Book: Horror Story Number 5, pp. 85-6 (and corresponding Trailer Cartoon on p. 88); indeed, it seems to me I just cited it the other week in a somewhat different context here on the List. But that isn't as gratifying as being able to point to a 14 (at least) year old RFC (or MAC Memo, I guess). Perhaps needless to say, I join you in urging all readers of this List to be particularly cautious not to cause squandering of others' cpu cycles, even if we can't prove just how long we've been preaching that particular position. cheers, map -------
PADLIPSKY@A.ISI.EDU (Michael Padlipsky) (08/11/87)
OH DEAR!!!!! I'm even more appalled to report on the resolution of yesterday's mystery about the "My NCP Costs $500 a Day" RFC: Through the good offices of Marlyn Johnson at the NIC, I've learned that there was indeed such an RFC (425, actually) but the reason I didn't spot it was that I was looking for ones by me and it (gasp, shudder, eek) wasn't by me after all. Could have sworn if was one of mine, but it turns out to have been by Bob Bressler--and addresses the general problem of too many Hosts having gotten into the habit even then (late '72) of doing "surveys," "probes," and the like. Oh, the forgetfulness of Early Middle Age.... abashed cheers, map -------
jkrey@AKAMAI.ISI.EDU (Joyce Reynolds) (08/11/87)
Dear Interested Onlookers, re: RFC 1012 and the "But my NCP costs $500 a day..." Please see page 28 in RFC 1012 for the RFC... 425 - Bressler, Bob, "But my NCP costs $500 a day..", RFC 425 (NIC 13010), BBN, 19 December 1972. Joyce Reynolds