[8257] in Commercialization & Privatization of the Internet
ANSnet backbone router problems??
daemon@ATHENA.MIT.EDU (Gordon Cook)
Fri Nov 12 12:10:54 1993
From: cook@path.net (Gordon Cook)
Date: Fri, 12 Nov 1993 17:09:53 GMT
To: com-priv@psi.com
A friend well to the west of Nearnet sent me this. I gather that the RS-6000
routers are being stressed in ways that they can't handle well and that the
stress is at bandwidths in the less than 15 megabit per second range. IE 1/3 of
what the NSF has been paying for for nearly 3 years now.
Could someone more technical than I offer their views on what is going on?
---------------
Retrieving information on NEARnet Trouble Ticket 5893. Please wait...
Ticket Number: 5893 Ticket Status: open
Ticket Type: unplanned Ticket Source: noc
Ticket Scope: network Site/Line: nsfnet-enss134
Ticket Owner: long Problem Fixer:
Telco Number:
Ticket Opened: 10/26/93 15:10 Problem Started: 10/26/93 15:00
Problem Description:
Seeing roughly 3 percent packet loss to the ENSS134 router.
Tests to other routers on the MIT external net run cleanly.
Have notified ANS Operations staff of the problem.
Note: 1 Last week we had a problem where the ethernet interface on
10/26/93 15:18 the ENSS134 router got wedged and required a restart to
perfetti fix it. Have reminded ANS staff of this and they will
now look into it and let us know what is found.
Note: 2 Nisar at ANS NOC reports that Eric Bennett is working on
10/26/93 17:53 this and from his last update 15 minutes ago, he has found
jhickey that a T960 card is feeding an Ethernet card packets faster
than it can handle them.
Note: 3 ANS NOC reports that the only way to fix the Ethernet buffer
10/26/93 22:04 problem is to offload some of the traffic from the Ethernet
jhickey to FDDI. Eric Bennett or Selina Priestly will be handling
this and our Eng. staff will need to coordinate with one of
them as to proceeding with preferring FDDI to the Ethernet.
Note: 4 Requested that ANS do an upgrade to enss134 rather than
10/27/93 14:02 offload traffic to fix the Ethernet buffer problem. Tony
hbryant from ANS is getting in touch with one of the engineers
and he'll have him get back to up.
Note: 5 Spoke with Tony and Eric at ANS. Eric reports that
10/27/93 14:03 router and it's ethernet card can not be upgraded, as
jbehrle they are both running on the latest hardware available.
Eric reports they are having a telephone confrence with
the folks at IBM today, and this will be a major item on
their agenda. Eric will call back after that confrence
(16:30ish).
Note: 6 Spoke with Eric and Tony at ANS, who report that they have
10/27/93 16:37 discussed this issue with IBM. The only solution that
jbehrle they are able to recommend is to off load some of the
traffic from the ethernet card, as it is unable to keep
up with the traffic from the T3 card, thus the router is
discarding packets. Eric said that someone from IBM is
willing to discuss this with us. Eric also noted that the
ethernet card is handling ~5Meg of traffic at any given
point in time.
Note: 9 Dan Long has spoken with Dennis Baron who reports that a
11/03/93 09:03 request has been put in for the fiber run to Tang, but he
csoulia has gotten no committment date from his wiring people.
Dennis expects it will be approximately 2 weeks.
Note: 10 John Oliveira of NEARnet will coordinate with MIT personnel
11/08/93 14:14 on the fiber run to Tang Hall.
perfetti
Note: 11 Several NEARnet members have reported 20% packet loss to the
11/10/93 15:37 enss today. We are working on offloading some traffic
long from that path immediately (rerouting MBONE tunnels) and are
planning to eliminate that path for outbound traffic tonight
between 8pm and 10pm. We assume that rerouting outbound
traffic to a router on the FDDI ring will suffice to prevent
the ENSS Ethernet interface from dropping packets. We still
plan to move all traffic (both inbound and outbound) to the
FDDI ring but are awaiting a fiber pull scheduled for 11/19.
Note: 12 There is a possibility of routing glitches between NEARnet
11/10/93 20:22 and the Internet from now until 23:00 tonight due to efforts
jhickey to alleviate the packet loss problem on the ENSS.
Note: 13 The ENSS went down at 20:20 in connection with efforts to
11/10/93 20:24 correct the packet loss problem.
jhickey
Note: 14 The ENSS ethernet connection was re-established at 21:26. It
11/10/93 23:18 had been accidentally dislodged in the course of actions
jhickey being taken to correct the packet loss problem.
Note: 15 Per NEARnet Eng. the default route at MIT routers will be
11/10/93 23:19 changed and there may be accompanying routing glitches.
jhickey
Note: 16 At 23:15 we have changed the way NEARnet routes outbound
11/10/93 23:45 to the NSFnet. We are currently routing outbound via a
skennedy connection to the FDDI ring that the ENSS also has a
presence on. Inbound traffic is using the usual NEARnet
connection to the ENSS ethernet. This means that until the
next NSFnet config window on Friday morning there will be
asymmetric routing. If the FDDI does go down, routing will
switch over. If the ethernet goes down routing gets even
stranger, but should work. Current tests show no packet loss
Note: 17 Testing indicates that our change to outbound traffic has
11/11/93 10:17 had a significant effect on NEARnet -> NSFnet performance.
skennedy We spoke with ANS, and they plan to have their side
configured on Friday at 0500 so that we can make routing
via the FDDI connection symmetric, and avoid the lossy
ENSS ethernet card.
For more information, please contact the NEARnet Network Operations Center
at nearnet-ops@nic.near.net or (617) 873-8730. Thanks!
