-------------------------- CONNECTION PROBLEMS UPDATE -------------------------- In the June 23 edition of BCPL System News I described the various types of connection problems being reported to the Help Desk and outlined the steps we had taken up to that point to diagnose those problems. There have been several developments since then, so it's time for an update. If you missed the June 23 message, you will find it in the System News archives on our Web pages. If you are using your UNIX shell account, simply enter "sysnews" at the UNIX prompt. If you are using a PPP connection, point your Web browser at http://www.bcpl.lib.md.us/sysnews.html. Login Failures & Unexpected Disconnects -------------------------------------- The consultant we brought in to help us has been working with Multitech and Cisco, going over our modem and terminal server configurations to see if there is anything we missed on the first round. So far they have found nothing wrong with the terminal servers, but are continuing to examine them. About two months ago we installed MultiTech's flash ROM upgrade version 2.13 in all of our MultiTech modems. Since then MultiTech's engineers have discovered a bug in their flasher utility that can corrupt the new ROM code if the connection speed between the computer running the flasher utility and the modem being upgraded is higher than 9600 bps. We used 34,400 bps when we did the upgrades, so there is a good chance that our MultiTechs now have corrupted ROM code. The solution is to use a "pre-flash" utility to completely wipe out the existing ROM code, then run the flasher utility at 9600 bps. We will begin this on Thursday August 1 and should be finished by Saturday August 3. MultiTech has also suggested several modem configuration changes, but recommends that we not implement them unless re-flashing the modems fails to improve connection reliability. The configuration changes may result in decreased throughput for users with 28.8 and 33.6 modems, so we will not implement those changes unless all else fails. One theory that we explored was the possibility that the network connecting our terminal servers, our gateway router, and our UNIX host was becoming overloaded with traffic, adversely effecting the performance of the terminal servers. We don't have any fancy network analysis equipment so couldn't verify that theory, but debug output from the terminal servers seemed to point in that direction. As an experiment, on Tuesday July 30 we installed a Cisco Catalyst 2800 Switch, with the UNIX host, the gateway router, and each terminal server connected to its own port on the switch. A switch acts like an electronic traffic controller to direct data packets from one port to another according to the origin and destination of each packet. Thus each packet is seen only by the destination device, not by every device on the network. Unfortunately, we discovered that this does not work well with high-traffic devices like our terminal servers. Although congestion on the network as a whole was reduced, congestion within the switch itself was such that communication between the router, terminal servers, and UNIX host was slowed down rather than speeded up. We removed the switch Wednesday afternoon. Ring - No Answer ---------------- Several times since the June 23 report we have found incoming lines that had gone dead, resulting in rings with no answer if you hit one of those lines. Today I found three more. Bell Atlantic is usually quick to repair dead lines, so those three will probably be back in operation by Friday. Busy Signals ------------ Busy signals are rearing their ugly heads again, especially during prime time. As temporarily relief, on Monday July 29 we added 16 modems, bringing our pool to a total of 176 modems. Additional modems will be in service by late August or early September. That's the news to date. If you have questions about this or about anything else pertaining to your BCPL Internet account, please contact the Help Desk by phone (887-3297) or e-mail (oz@mail.bcpl.lib.md.us).