@make[report] @device[x2700] @flushleft[Edinburgh Regional Computing Centre] @majorheading[Report on Pilot TelePAC PSE] SEEL made a final delivery of hardware on the 29th April that completed the 30 port, 2Mbyte TelePAC switch. The software that was delivered with the switch was version 0468 and the testing has been done using that version. The switch has been in user service since the 8th April and the load has been built up steadily. There are currently 19 active connections with a typical loading of nearly 100 calls. @heading[Hardware Report] The first part of the switch was delivered on the 28th March, during the initial testing it became clear that this unit had a faulty backplane. SEEL turned out very promptly to fix it but the obscure nature of the fault meant that the unit was down for about 16 hours. Since that incident there has been no other hardware failures with the unit. One problem has emerged when using lines timed by the TelePAC. CRC errors are observed on the line and in the worst cases some faulty blocks are passed as correct by the CRC check. On investigation this would appear to be caused by using cables that are too long although the situation was confused by widely differing margins in particular V24 interface boards. The better boards will run cables of up to 40 metres in length with no problems at 9.6kb/s. The worst boards start to have problems at around 25 metres. This means that the boards are strictly within spec (50 feet) but SEEL and Telefile are looking at the problem and expect to find a solution. @heading[Software] The meeting between the JNT, Telefile, SEEL, Cambridge and ERCC on the 5th Feb. 1986 listed a number of software problems and deficiencies. These are itemized below along with a report on which have been fixed with the current software release. A number of other software faults have been discovered during the evaluation and they are listed at the end. @subheading[Guaranteed PSS compatibility (BG 1) ] SEEL have guaranteed that PSS compatibility will be maintained. @subheading[Call statistics in Clear packet (BG 2) ] PSS style call statistics have been added as an option on a line by line basis but they are currently only working on incoming call clear packets not clear confirmation packets. This has been logged as a bug (see N7). @subheading[Call redirection (BG 3) ] Call redirection now works although the method used to do it is a bit obscure. @subheading[Load Sharing (BG 4) ] Load sharing has been included in the delivered version of software, it has been checked and the calls are shared across the lines in the load share but when one of the lines in a load share fails and then restarts, it is not re-used. This has been booked as a new bug - see N6. @subheading[Performance monitor (BG 6)] Various global statistics, such as the total number of calls, current switch packet throughput, number of rejects and retransmissions can now be listed out on the switch operator console or remote operator console. @subheading[Fault warning (BG 7)] A new message - error threshold exceeded - has been included to warn of excessive number of errors. Currently the number of CRCs, Rejects and retransmissions are counted. @subheading[Network management (BG 8)] With the addition of the new counts the switch now meets the spec for single PSEs. I have not tried out the management package for multiple PSEs. @subheading[Various packet counts (BG 9)] The following packet types are counted and may be displayed:- Calls Transmitted, Calls Received, Clears, Interrupts, Resets, Data and CRC errors but Clear, Interrupt and Reset packets transmitted by the TelePAC are still not counted. @subheading[Displayable DTE addresses (BG 11)] The DTE addresses of a call can be displayed on a line by line basis, the ability to list calls by address has not yet been provided although it is expected in the next release of the software. @subheading[Password on BASE: (BG 12)] There is now a password that the operator may change to control access to the remote operator facilities. @subheading[MTBF (BG 13)] SEEL have guaranteed a 2000 hours MTBF. @subheading[Crash on buffer exhaustion (BG 14)] The switch will take action to prevent a crash when the number of buffers goes below a critical level. The value for this critical level can be set by an operator although the change will not come into effect until the next re-boot of the software. @subheading[Hold link down (BG 15)] It is now possible to 'STOP' a line to hold it out of service. @subheading[Change of line speeds (BG 16)] It is possible to change line speeds without re-booting the switch software. @subheading[Reset code of 0/0 on errors (BG 3)] The TelePAC will now generate resets with non-zero network clearing codes on lines designated as DCEs. @subheading[Clear code of 0/0 on link down (BG 4)] The TelePAC now returns the code 9/0 on a line that is down. It has been pointed out to them that 9/43 is preferable when working in the current GEC multi-switch environment. @subheading[Packet size of 256 (BG 13, CJC 3)] The current version of the software still does not support full 256/256 working. Various faults in the software concerning negotiation of packet sizes have been fixed so that it will properly interwork with GEC switches. The current position is that if a DTE requests 256/256, the DTE is allowed to transmit 256 byte packets but the TelePAC will offer a negotiation of 128/128 to the remote end and then fragments the 256 byte packet using the 'M' bit so that the packet sequence is legal for the remote end (or intermediate switches). @subheading[No response to REJ command (CJC 1)] This has not been fixed. @subheading[Spurious pairs of Clears (CJC 2)] The fault which generated more than one restart packet has been cleared. @subheading[Crash because of incorrect configuration (CJC 4)] It is still possible to cause an immediate crash of the TelePAC by setting 'wild' values during the configuration. The situation has been considerably improved and the SEEL manual ought to lay down the possible values for the parameters. @subheading[Delays on response to Call Request (BG 9, CJC 5a, 13)] This problem has now been fixed although it is recommended that a 'catch-all' address is added at the bottom of the X.25 address table to ensure that a match is always made to an incoming address to avoid the call being routed to an internal resource. @subheading[Wrong LCN used after Clear to Call Request (CJC 5b)] This fault is cleared. @subheading[Clear held up until after Accept (CJC 6b) ?] This fault has not been cleared. @subheading[Reset is not propagated (CJC 7) No] Resets are now propagated through the switch. @heading[Desired Enhancements] At the meeting of the 5th Feb. a number of items were put on a 'desired' list, these items are listed here for completeness. @itemize[ Unmatched called addresses (BG 5, CJC 12) No response to illegal LCGN (BG 6) DTE sends network clearing codes (BG 7) Display of link output queue size (BG 8) Change number of LCNs should not require reboot (BG 10) Documentation (BG 11, CJC 22) Respond Clear Conf. to Clear after Call Req. (CJC 6a) Need multiple LCGs (CJC 9) Supervisory command recovery after REJ (CJC 10) Lower and upper case (CJC 11a) DEL (CJC 11b) Non-immediate configuration (CJC 15) Selection of address table entries (CJC 16) Configuration of address table (CJC 17) DTE/DCE instead of A/B (CJC 19) Physical DCE presentation (CJC 20) X.25 subscription options (CJC 21) Addresses in billing records are zero-filled (CJC 8) ] @heading[New Faults Discovered during the Trial Period] @itemize[ N1 When a call is 'timed out', the TelePAC repeats the call attempt 5 times. It should not repeat the call attempt. N2 It appears that when line faults are occuring, some bad packets can get through the CRC check & this causes then provokes the TelePAC to disconnect the link. N3 Configuration. If a large number of Billing records are asked for (but less than the stated maximum), the TelePAC crashes on BOOT & it can be very difficult to stop it looping in the diagnostics. N4 After the above, the only way of restoring the TelePAC is to reset the memory, it is NOT enough to simply reset the number of billing records and re-boot N5 Buffers dropped from almost 400 to 173 for no apparent reason. The number stayed that from 5pm to 1pm the next day. It then rose again to 260, again for no apparent reason. N6 Load Sharing. When a load shared line has gone down and up, calls do not appear to then use that line. N7 PSS Statistics. Only operate one way round. N8 When a line attached to a line driver goes down, the switch puts out a constant stream of messages in the form "Input too long". This is thought to be caused by the switch not searching for a flag packet after the 1st error. N9 When a call is negotiated with a packet size of 128 and a packet > 128 is sent to the TelePAC, a) it is not reset and b) the oversize packet is sent to the receiving host. (version 0451) N10 When a call is sent into the TelePAC with a null called address, the call is routed internally within the TelePAC, this causes, from the user's point of view, a delay of one minute before the call is cleared with a code of 0/0. ] @heading[Summary] The PSE has now been in user service for a number of weeks and is basically sound. It is an acceptable switch for the Academic Community but it will be necessary to continue to pressure Telefile to fix the remaining faults and start work on the 'desired enhancements'. @flushright[B.Gilmore May 1986]