@make[report] @flushleft[Edinburgh Regional Computing Centre Communications Group] @majorheading[X.25 Packet Switch Survey] @heading[2nd Stage Report] @Subheading[General] After the conclusion of the first stage survey, it was agreed that 6 companies be examined in more detail, they are Amdahl, Camtec, Dynatech, GEC, Telefile and Telematics. In view of the position of GEC and Camtec as existing suppliers in the Academic Community, it was not considered necessary to examine their equipment to the same depth as the rest. Amdahl, GEC and Telematics are considered as suppliers for the high connectivity and high throughput switch requirement while Camtec, Dynatech and Telefile were considered for the small switch needs. It should be noted that all costs are ex VAT @subheading[Basis of Comparison] Each manufacturer was given a copy of the JNT X.25 Campus Packet Switching Exchange Requirements along with a document (Annex 1) which listed a further number of requirements and two additional configurations, for a 10 line and a 100 line system. The specification and additional points raised by the JNT were discussed in full with the manufacturers and a detailed report is given for each company later in the report. @subheading[Results of the Comparison] @paragraph[High End] It does not appear that GEC will be able to provide either a 100 line switch, or the desired throughput, before the third quarter of 1986 at the earliest. Even at this date is is not clear that it would achieve the 1000 packets/sec throughput. Both Telematics and Amdahl can provide large switches. Amdahl is considerably more expensive (~40%) and a number of its facilities, particularly Network Management, are more basic than Telematics. Of the three, Amdahl has a much better record of reliability than the others with GEC looking an extremely poor third. One disappointing fact is the current lack of high speed lines, except from GEC. @paragraph[Low End] The Community has been waiting for the CAMTEC switch pad for a long period. The first version has appeared for field trial but it will be January 1986, at least, before a usable switchpad appears. When measured against the specification it lacks a large number of features. The only real advantage it shows is the extremely cheap price. Dynatech produce a very disappointing product for such a 'main stream' company. The Telefile product, as tested, is not perfect but does perform as claimed. There are a number of points where it does not match the JNT spec but Telefile are currently acting to rectify these as they appreciate that this will assist them in the overall market place. The cost of the Telefile product is higher than Camtec but still represents excellent value for money. Subject to the modifications and clarifications mentioned in the individual reports being progressed satisfactorily, the recommendations of this survey are:- @subheading[Summary] @paragraph[High End] Telematics offer the best value for money for this size of switch, provided that software license and support issues can be simplified for the Academic Community. @paragraph[Mid Range] It would seem reasonable to allow open tendering between GEC and Telematics for the mid range products. @paragraph[Low End] The Academic Community should be able to choose either Camtec or Telefile for small switches. @flushright[B. Gilmore December 1985] @set[section=0] @newpage @majorheading[Amdahl] @section[General] Amdahl supply a wide range of X.25 equipment which can be put together to provide extremely large networks. The equipment is based on a multimicroprocessor architecture and although no store sizes are quoted it is stated that the largest switch can handle 2000 simultaneous calls, expandable to 6000 calls. @section[Hardware Description] The main Amdahl switch, called the 4410 Network Processor, is based on a multi microprocessor design using Computer Automation microprocessors. Each X.25 line (or pair of low speed lines) is controlled by an I/O Module containing a CA micro and 64K bytes of local memory. The I/O Module handles all of both Levels 1 and 2. The I/O Modules are connected to a Data Movement Processor which interfaces to the main Packet Level Processor (PLP). The PLP has private memory for its code and access to 128K bytes of public memory for buffers etc. This entire unit is called an X.25 processor (XPRO) and the largest switch configuration contains three. There is a further Node Administrator, a CA mini, to control the parts and to handle all the details of call setup and cleardown. It is an unusual configuration as there is only a small amount of global memory, with most of the memory (in excess of 7M bytes in the biggest switches) being held in the line modules. @section[Software Description] All the software is written in assembler. It is claimed to be modular and cannot be amended by the 'user'. @section[Configuration Options] The smallest switch consists of a single XPRO and Node Administrator and can be expanded up to 40 lines. The addition of a second XPRO adds a further 40 links and the third XPRO takes it up to the maximum configuration of 120 lines. @section[Line Speeds/Interface Options] The following line speeds are supported for both internal and external clocking: @begin[verbatim] 2.4k, 4.8k, 9.6k, 19.2k, 48k, 64k. @end[verbatim] V24 is used for speeds up to, and including, 19.2k, V35 is used for 48k and 64k, it was suggested that the way to handle higher speeds was to use a number of 64 kb/s lines using an Amdahl 2210 multiplexor. @section[Throughput] A single XPRO can handle 450 packets/sec (all packets through the Node), a fully configured Node can handle 1350 packets/sec. There is also an aggregate data limit of 300 kb/s full duplex, rising to 900 kb/s for the fully configured node. @section[Costs] @begin[verbatim] 48 ports @T{#}158,000 100 ports @T{#}240,000 @end[verbatim] Additional lines are @T{#}1,000 per 9.6 kb/s and @T{#}2,000 per V35 line. The significant price jumps occur at 40 and 80 links. @section[X.25 Facilities] Fast Select, Packet and Window negotiation, PVCs and Reverse Charging are supported. Call Statistics are not currently supported in a PSS compatible form. There are no plans to support Call Redirection. The maximum packet size supported is 256 bytes. @section[Multinode Networks] Amdahl switches are currently used in large multinode networks (AT & T have 50). A modified form of X.25 is used between the nodes and although the protocol is not in the public domain, Amdahl would be prepared to release it to the Academic Community. Adaptive routing is supported. @section[Addressing] A single address is normally used per link, but 'Access Groups' may be used to achieve the desired flexibility. An Access Group is normally used to create a 'hunt group' effect, in other words it specifies a single address, or range of addresses (created by an 'n' digit match) followed by a list of lines corresponding to that address (or range). Normally the number of lines would be greater than one but by restricting it to one line and having a number of access groups all pointing to that line, the effect of multiple ranges and single addresses is created for a particular link. Full load sharing is not supported. If multiple links correspond to an address or range then a 'round robin' system is used to allocate calls to the links. @section[Network Management] Each node in a network has up to three asynchronous ports for control purposes. The first port is used to issue commands to the processor, the second for monitoring the performance and printing alarms, the third for the output of billing records. The functions of any of the ports may be combined with those of another so that only one port is mandatory. In order to perform Network Management of a Network, these ports are connected into reverse PADs (with an Amdahl PAD, a PVC would then be used) and the PADs connected to a remote management machine. Software currently exists for using an Amdahl mainframe running UTS as the management machine. It can be seen therefore that "simple" X.29 is used as the protocol and full details exist to allow the Community to use its own host machines as desired. @section[Statistics] Statistics are collected as required by the spec, however, it is not currently possible to store them on a disc on the switch. There is a development in hand to attach a 40 Mbyte Winchester (with floppy access) to a 4410, to be called a 'Z-Disc', which will then allow the collection of statistics. @section[Operator Control] The details noted in the section on network management apply to control although the functions can be split as noted above. It is not possible to have both local control and remote control simultaneously. The various aspects of control as specified are possible on a live system with the exception of the ability to clear a call to a specified address. @section[Date-Time Stamp] Messages are date and time stamped. If a node crashes, and it is a single node system, then the current date and time are lost although the date and time may be changed on a running system. On a multiple node system, the date and time are automatically recovered. @section[PSS Compatibility] Amdahl do not guarantee compatibility with PSS. They claim to 'conform to the spirit' of the CCITT recommendations. In reality, they claim that they would amend their software if we discovered differences on our existing machines. @section[Performance - Packet Transmissions] @begin[verbatim] Call Requests 80ms Data Packets <10ms Clear Request 80ms @end[verbatim] Although the Call Request time is rather long, it is handled independantly of the main switching function. @section[Availability and Reliability] The switches have been designed with the aim of high availability. There are, for example, three power supplies on each node, if one fails it can be replaced without stopping the system. W. H. Smith has 4 and are quoted as saying 'they can't remember when they last failed'. An I/O Module can also be replaced on a live system. If an entire XPRO goes down, then it can be repaired without a reload of the entire node. Obviously if a Node Administrator, or its associated bus, fails then the entire node fails. @section[Operation] The system has been designed to run unattended and meets the spec. @section[Reconfiguration] When a system is first installed, Amdahl will have set up an initial configuration. It is then possible to change any of the configuration details, or set up new lines, using the built-in management software. @section[Known Problems] @begin[verbatim] No call statistics. No guarantee of conformity to PSS. No full load sharing. No storage of call stats (currently). No list/clear of calls to a specific DTE address. @end[verbatim] @section[X.25 (1984)] Amdahl are aware of both the X.25 (1984) specification and the ISO DP8878 document. There is not a timescale currently for the implementation of the standards. They are prepared to discuss the implementation of DP8878 when an order is placed by the Community. @section[Academic Community Special Requirements] They would be prepared to discuss the implementation of specials for the Community, although they would prefer not to integrate such software into a 4410. The preferred vehicle would be the hardware used in the latest Amdahl PADs. @section[Product History] The 4410 has been available since 1980/81 although there have been several hardware upgrades since that time. The total sales in the UK have been of 12 units, the customers including W.H. Smith and ICL. The sales worldwide have been about 100 units and include one PTT (South Africa). @section[Maintenance Arrangements] The maintenance costs are 10% of capital for 'complex' items (those with discs) and 8% for the rest. The cover would normally be 9-5 Monday to Friday excluding public holidays. The current call out period varies from 2 hours to 6 hours (soon to be reduced to 4 hours) and varies depending on the locality and is subject to negotiation with the Product Support and Services division. There is an established escalation procedure starting after the engineer has been on site for one hour. @section[Company Information] Amdahl is a large US owned company (with Fujitsu holding 46%). All software and hardware development is done in the USA (headquarters now in Dallas). There are 10 people in technical support in the UK. @section[Confidentiality] None. @section[Contact Name] @flushleft[T. W. Arnold Amdahl House 112/116 Cricklade Road Swindon Wiltshire SN2 6AG Telephone No 0783 45476] @section[Summary] There are a number of points where Amdahl fail to meet the JNT spec. In particular the Network Management facilities are surprisingly sparse. The claims for availability and reliability are more credible than the competitors, particularly GEC but as the cost of the switches is greater than the other short listed competitors, Amdahl do not represent very good value for money for the Community. @set[section=0] @newpage @majorheading[Camtec] @section[General] In view of the position of CAMTEC as suppliers in the Academic Community this report is in the form of an update only. @section[Hardware Description] The CAMTEC switchpad contains a new board consisting of a M68000 processor, 1/4 megabyte store, 4 synchronous line interfaces and a Ring interface. Either one or two of these boards can be combined with the Z80 processor board and 128k bytes of store found in a standard PAD. It is expected that eventually 4 M68000 boards could be used. A CAMTEC Ring Switch consists of up to 6 Ring Gateways interconnected by a dual Cambridge Ring. In this configuration the line speed is constrained to the 19.2kb/s of the Ring Gateway. The code in the Ring Gateway has been modified so that X.25 packets can be carried across the Ring in a non Ring-standard fashion. There will be a further development that uses switchpads interconnected by Cambridge Ring but this will not be available before the summer of 1986. @section[Software Description] The Z80 is being used to move packets between the M68000 boards. Only packets that need to be switched to another M68000 board are handled by the Z80. On the current releases of software the Z80 handles all call setups and clear downs although this should change in the longer term. @section[Configuration Options] On a single M68000 card, 2 of the lines can run at 64kb/s, the other two can be currently run at speeds up to 19.2kb/s and it is hoped that they will run at up to 38.4kb/s in the first quarter of 1986. @section[Line Speeds/Interface Options] V24 is used for speeds up to 19.2kb/s, X.21 is used for higher speeds. V35 is not supported although a converter between X.21 and V35 is available (not from CAMTEC). Internal speeds of 2.4, 4.8, 9.6, 19.2, 38.4, 48 and 61K are supported. @section[Throughput] A throughput of 100 data packets/sec is claimed for a M68000 board. The Z80 will be a limitation on the throughput of an entire switch, although it is claimed that a Z80 could handle 59000 bytes/sec in the later versions of the switch when the Z80 is relieved of other duties. @section[Costs] The following costs are given in two parts, first the cheapest way to put the unit together, ie. using existing PAD channel boards to provide the 9.6kb/s lines and the second using all 68000 cards with a minimum line speed of 19.2kb/s. The costs are 'Academic Community' prices. @begin[verbatim] 10 Lines JNT PAD with single upgrade @T{#}4,500 Full switch box, ie. 3 x 68000 cards @T{#}8,240 18 Lines Single Switchpad with 10 Z80 lines, 8 M68000 lines @T{#}7,500 Switch 16 + 4 ports Minipad (budgetary price) @T{#}11,9000 @end[verbatim] @I{N.B. Both options for 18 lines require a Switchpad with 4 M68000 line cards in it rather than the 2 existing cards.} There are plans to create switches to meet the 48 line or 100 line switches but there is no possibility of them before late 1986. The costs have been put in annex II to the report. @section[X.25 Facilities] Fast Select, Packet & Window size negotiation, Call Redirection and Reverse Charging are supported. Call Statistics are not supported although CAMTEC are 'now aware of the problem'. CAMTEC are now committed in a contract to implement PVCs. The current maximum window size is 256 bytes, it is hoped to increase this to 512 bytes by the third quarter of 1986. @section{Multinode Networks} CAMTEC use 'simple' X.25 between network nodes. There is uncertainty as to whether Network Clearing Codes can be passed. No special action is taken to retain calls if a transit node, or its links, break. @section[Addressing] CAMTEC use an address match scheme to construct address tables. There can be single addresses, 'ranges' or mixtures of both types. The 'ranges' mechanism is restricted to match on the first 'n' digits and the common failing of this scheme, ie. address 1234 is not the same as 000000001234, applies. No full load sharing exists. @section[Network Management] The current network management in the switchpad is extremely limited. There is no data collected for each call, there is a very limited performance measurement and there is no warning of fault conditions. There are plans to provide such information but it is in Release 4 of the software, not expected before mid 1986 at the earliest. When it is released it will use a DEC Micro VAX II. There are plans to use an IBM PC but that will be even later. A special 'block' protocol will be used and CAMTEC are prepared to put this protocol in the public domain but not before the end of 1986. @section[Statistics] Only counts of data packets are provided. Any more facilities will await Release 4, ie. not before mid 1986. The current plans are not to use a local console at all so all collection will be have to be performed remotely. @section[Operator Control] It is not possible to enable or disable a link or clear a call to a selected DTE address. It is not possible to change line speeds, LCNs or the DCE/DTE sense on a running system. These features are again promised in Release 4 or even the second version of Release 4. @section[Date-Time Stamp] Not before Release 4. The NMS will then hold the time and correct the time in PADs and switches. @section[PSS Compatibility] Compatibility is committed. @section[Performance - Packet Transmissions] The following times were measured on a 6Mhz Z80 for call setup. @begin[verbatim] Call request 61.6ms Data Packet 5.2ms Clear Request 5.2ms @end[verbatim] It seems unlikely that a switchpad will do better on Call Requests as it will still use the 6Mhz PAD. @section[Availability and Reliability] The hardware is claimed to have an MTBF of more than 2000 hours, Camtec did not seem to be prepared to make any guarantees on the software MTBF for single switchpads. @section[Operation] Complies, apart from the ability to dump. @section[Reconfiguration] Complies. @section[Known Problems] The switchpad, even in its first full release scheduled for January 1986 does not meet the spec in a whole number of points. It remains to be seen to what extent the Release 4 software will, and what the timescale for its release will be. @section[X.25 (1984)] Implementation of the ISO Network Service has been committed for the MSC network by the end of 1986. There is a realisation that options will be required for the various versions. @section[Academic Community Special Requirements] Prepared to consider. @section[Product History] Known. @section[Maintenance Arrangements] As before, switchpad cost not requested. @section[Company Information] Known. @section[Contact Name] Known. @section[Confidentiality] CAMTEC is sensitive about the details of the backplane, for example the constraints caused by using the Z80 processor. @section[Summary] If this switch was tendered by any other supplier it would probably not be accepted in its current state of development. However, since they are, to a certain extent, just an extension of the current PADs, and are extremely good value for money it does not seem practicable to stop their use. @set[section=0] @newpage @Majorheading[DYNATECH] @section[General] Dynatech sell a range of X.25 equipment including PADs, switches and a Gateway Address Translator and a Network Control Centre. Dynatech sell two switches, a low maximum line speed and a high maximum line speed. Only the second one has been considered for the second stage report. @section[Hardware Description] The switch, called the Multi switch X.25, is based on a single 8 MHz Z80 processor with a total of 248 Kbytes RAM. The basic switch consists of a box with power supply, a CPU board and two trunk boards each capable of handling two lines. A further 4 trunk boards can be installed in the box. @section[Software Description] Z80 assembler is used throughout. @section[Configuration Options] The Model 12 comes with 4 ports, a further 8 ports may be added in pairs. @section[Line Speeds/Interface Options] The following internal clock rates are supported: @begin[verbatim] 1200, 2400, 4800, 9600, 19.2k and 57.6k @end[verbatim] Any line may be externally clocked up to 64 kb/s. V24, V35 and X.21 are supported. @section[Throughput] A minimum throughput of 100 data packets/sec through the PSE is claimed. @section[Costs] New costs have recently been quoted, partly brought about by a marketing rethink and partly as special prices for the Academic Community. They are:- @begin[verbatim] 4 links @T{#}10,006 6 links @T{#}11,345 8 links @T{#}12,684 10 links @T{#}14,024 12 links @T{#}15,364 @end[verbatim] @section[X.25 Facilities] Call statistics (as per PSS) are not provided. Packet and window negotiation is apparently supported but regardless of the values negotiated a call always gets the maximum figures that the switch is set up for. Packet sizes of 128 and 265 are 'supported'. A maximum of 127 logical calls per link are supported. @section[Multinode Networks] Dynatech use 'simple X.25' to communicate between multiple nodes on a network. There is a special marker on a trunk to indicate that the trunk, acting as a DCE, is connected to another switch and not to a simple DTE to allow network clearing codes to be passed between switches. Trunks may also be grouped together to allow load sharing (based on the smallest number of active calls) on multiple connections. @section[Addressing] A maximum of 16 entries is allowed for each trunk in the routing table. Single addresses as well as address with 'don't care' digits are allowed. A trunk may be either a DTE or a DCE although it is controlled by a strapping on the CPU board and therefore the switch must be powered off to change the setting. @section[Network Management] Dynatech use a software package on an IBM XT to create a Network Control Centre (NCC). The NCC can poll a number of switches to read off statistics, note alarm conditions etc. It does not appear that detailed statistics are held for each call and as the machine is polling, alarms will only be noted at the poll rate which are from 1 hour to 4 hours in steps of half hours. The information when collected is stored on the PC's disc for late display or printing. The NCC uses a simple character stream running over X.29 to collect the information. For example, it would issue 'STAT 5' to receive link status for Line 5 in the same manner as an operator on the switch control console. This could be emulated by another machine. @section[Statistics] The Model 12 only collects information on - @begin[verbatim] number of Resets number of Data Packets number of CRC errors @end[verbatim] The statistics may be stored in the battery backed RAM to avoid loss. @section[Operator Control] Claimed to meet the spec - but certain features, such as clear a call to a particular DTE address, will only be available from the NCC. @section[Date-Time Stamp] Not performed by the switch but can be provided by the NCC. @section[PSS Compatibility] Complies. @section[Performance - Packet Transmissions] @begin[verbatim] Call Request 11-20 ms Data <10 ms Clear 15 ms @end[verbatim] @section[Availability and Reliability] Claimed to meet the spec. The hardware MTBF is calculated at 15,000 - 20,000 hours, experience with the equipment in the field is claimed to backup that figure. It is also claimed that there have not been any software crashes in the units installed in the UK. @section[Operation] Complies; software is contained on EPROM. @section[Reconfiguration] Configuration is performed 'on-line' or can be held in the NCC. The system needs to be re-booted to change either:- @begin[verbatim] a link from DTE to a DCE or Internal clock to external clock. @end[verbatim] @section[Known Problems] @begin[itemize] Although it is claimed that the window and packet negotiation will work, in reality a call always gets the maximum window and packet size that the switch is configured for even though the negotiation has apparently worked. The switch does not appear to have enough buffer space, it was easy to provoke it into putting RNR up with only 60 calls in use. As a DCE it doesn't poll after receiving an RNR from the DTE. When a line is configured as a DCE it does not poll to start the link up. This is not compatible with PSS and would cause problems with existing DTE implementations. A line is immediately restarted on losing control signals. This prevents the insertion of a line monitor in a link. @end[itemize] Dynatech claim that the latter two faults were caused by them inadvertantly demonstrating a version prepared for use on Transpac and that the 'PSS version' would not do this. @section[X.25 (1984)] Dynatech say the are actively working on it and hope to have it ready for trial in early 1986. @section[Academic Community Special Requirements] There is a separate U.K. Software company that would be interested in doing converters and other special requirements for the community. The preferred vehicle would be the Dynatech 'MultiCom X.25' product which has a rather restrictive 9.6 kb/s maximum line speed. @section[Product History] The Model 12 was launched in the UK in November '84, there are currently 5 customers in the UK with twelve installed units. An additional order for 15 units has been placed by one of the customers. It is estimated that 500 to 1000 have been sold worldwide. @section[Maintenance Arrangements] It was suggested that a third party would be approached by Dynatech for maintenance if the Academic Community bought the product. The basic cost would be 10% of capital. If the customer chose to board swap and return then a 'special rate' would be quoted for spare boards. The maintenance cost includes both hardware and fixing any bugs in the software. New releases of software cost about @T{#}150 for the set of EPROMS which may be bought if desired. @section[Company Information] Some development is done in the UK, in particular the 'Protopad' and 'Monopad' but all model 12 development is done in the USA. The Network Control Centre software is being done in the UK. The development team in the UK is 3 people. @section[Confidentiality] None. @section[Contact Name] @flushleft[Dynatech Communications Ltd., Fieldings Road, Cheshunt, Waltham Cross, Hants, EN8 9YN. Telephone No. 0992 33555.] @flushleft[Northern Sales Rep: Keith Southerton, Telephone No. 0533 402807.] @section[Summary] There are a number of problems with the Dynatech Model 12. In addition to this, the quantity of store available for buffers does not seem to be adequate for a 12 line switch. Previously its overall price was extremely high but is now more in line with the competition. @set[section=0] @newpage @majorheading[GEC] @section[General] In view of the position of GEC as suppliers to Academic Community, this report is an update only. @section[Hardware Description] There are two major developments due for GEC switches, the MCC - megastream controller card and the KCC - kilostream controller card. Both cards use an on-board M68000 as a processor. The MCC is currently being field trialled in METSATNET, the KCC should be field trialled in the first quarter of 1986. In both the MCC and the KCC, all of both levels 1 and 2 of X.25 are handled on the M68000 card. @section[Software Description] In order to increase the number of lines on a 4190 beyond 60, or to run either the MCC or KCC, it will be necessary to run the Type 3 software. This software is under field trial currently, the first full release will be in the 1st quarter of 1986. This version will not include the 'JNT enhancements', which should appear in the 2nd quarter of 1986. @section[Configuration Options] The first release of Type 3 will not allow a 100 line switch to be constructed. The maximum currently envisaged is 99 lines and it will take about 3 months of extra software effort to achieve this. The hardware can be constructed by adding a 3rd FMC onto a 4190. @section[Line Speeds/Interface Options] The MCC supports G703. By the 1st quarter of 1986, both the MCC and KCC will offer X.21 (leased line). It is not planned to support X.21 on the existing PCC boards. The MCC supports a single line at up to 2Mb/s. Up to 2 MCCs can be supported on a single DMAD (DMA Director). The KCC will support either 4 lines at 64kb/s, 2 lines at 128 Mb/s or a single line at 256kb/s. Up to 7 KCCs can be supported on a single DMAD (although this hasn't been tested). One important advantage of both the KCC and MCC is that they do not require 'way numbers' and consequently can be mounted in the empty slots of an FMC that already has the maximum complement of PCC boards. @section[Throughput] The handling of X.25 level 2 on the KCC and MCC card will take processing load off the main switch CPU. This will mean an overall higher performance for the switch. Unfortunately, GEC are not yet prepared to quote a figure for this although it is suggested that a switch constructed entirely of the new cards would have a throughput nearly double that of an existing processor. A 99 line switch could therefore reach a performance of 1000 packets/sec, but would need a substantial number of its lines handled by the new cards. @section[Costs] @begin[verbatim] 4160 with 18 lines: Full Price - @T{#}36,000 With educational discount - @T{#}29,000 4190 with 48 lines: Full Price - @T{#}85,000 With educational discount - @T{#}66,000 4190 with 99 lines: Full price - @T{#}128,700 With educational discount - @T{#}107,200 @end[verbatim] @I{NOTE: It will not be possible to produce a 100 line switch in 1986} The additional costs for adding KCC lines are: @begin[verbatim] The DMAD: @T{#}3,500 KCC: @T{#}4,480 (inc. I/Fs) @end[verbatim] @section[X.25 Facilities] All. @section[Multinode Networks] Under Type 2 of the software, there are problems with passing 'Clear Requests' with a 'Network Cause' between GEC switches in a multinode network. Even under Type 3 software there will be problems unless all the nodes are under the control of a single node or a Network Management Machine. @section[Addressing] Complies. @section[Network Management] Enhanced facilities are being promised but certainly not before the 3rd quarter of 1986. @section[Statistics] Complies. @section[Operator Control] Enhanced facilities are being promised, in particular all DTE tables will be 'soft' and all the DCE tables, with the exception of the rather critical ability to change routing tables (which rather negates the advantage) but again not before the 3rd quarter of 1986. @section[Date-Time Stamp] Requires the 'JNT enhancements'. @section[PSS Compatibility] Complies. @section[Performance - Packet Transmissions] Complies. @section[Availability and Reliability] At this stage, there are no promises that the 48 line switch, far less the 99 line switch, will reach an MTBF of 2000 hours. The software reliability is liable to get worse when the Type 3 software is introduced until it has been in service for a considerable period of time. @section[Operation] Complies. @section[Reconfiguration] Complies. @section[Known Problems] The major problems are that even a 99 line switch running Type 3 software with the JNT enhancements will not be available until the 2nd quarter 1986 at the earliest and that its reliability will not meet the 2000 hour MTBF requirement. @section[X.25 (1984)] Marketing stated long timescales on the availability of X.25 (1984), however, Engineering recognise the desirability of providing the ISO Network Service and within the past few weeks have allocated a person to estimate the amount of effort required. Until this exercise is complete there are no timescales for implementation. @section[Academic Community Special Requirements] GEC would be prepared to do the 'special' converters for the Academic Community though obviously the question of realistic timescales would have to be entered into very carefully. @section[Product History] Known. @section[Maintenance Arrangements] As before. @section[Company Information] Known. @section[Contact Name] Known. @section[Confidentiality] None. @section[Summary] Although GEC are taking a more realistic attitude to their customers of late, it does not appear that they can offer a 100 line switch in less than 6 to 9 months. Even when such a switch is marketed there will be doubts about its reliability. @set[section=0] @newpage @majorheading[Telefile] @section[General] Telefile is a small company (owned by a U.S. parent) producing a switch called TelePAC. The switche's origins lie in a stat multiplexer switch to which X.25 switching capability was added. The UK company handles all the hardware and software development of the switch. TelePAC units will be able to be joined together using a DMA interface. This interface could also be used to connect to a user-programmed processor supporting a UNIX-compatible system for use as a Gateway or other such protocol converter. SEEL Ltd., Livingstone, now have the sole rights for sales in the Academic Community. @section[Hardware Description] The TelePAC is based on the standard VME bus with an 8 MHz M68000 processor. A variety of card cages can be supplied which then dictates the maximum configuration for that particular switch. The software for the switch is supplied on EPROMs, a card containing 64 Kbytes of battery backed RAM is used to contain the configuration and statistics. The switch that has been tested at ERCC had a further two 64 Kbytes cards installed. Versions produced now can also contain a 2 Mb store board. Further options for 1 and 4 Megabytes will be available very shortly. The current model of the switch does not need or support any disc storage, a floppy disc is being added as an option to allow more permanent storage for the configuration data and also to allow the switch to be dumped. An X.25 link is supported on an interface which contains the driver chips and clock source. Each pair of X.25 links are controlled by a DMA board which contains a Motorola DMA chip that can handle 4 concurrent 'channels' into main memory. All of level 2 is handled by the 68000. By the end of January 1986, a version of the switch using the DMA to connect directly two or more boxes together will be available. The current position is that the hardware has been connected but no software support is available. @section[Software Description] The software for the switch has been written in M68000 assembler, there is a reasonable division between the various parts of the software. From testing, it is very clear that the code has been written extremely carefully to optimise and sustain the packet throughput of the switch. @section[Configuration Options] An X.25 line is plugged into an interface card, which comes in several types to support the different interfaces. Two interface cards are supported by a single DMA card. The size of card cage used determines the maximum number of lines, the smallest will take 5 DMA cards and hence supports 10 lines. A larger card cage is also on offer than can take up to 16 DMA cards and hence 32 lines. This cage is used to build standard configuration of 8, 12, 16 lines all of which can be expanded up to the maximum 32 lines. @section[Line Speeds/Interface Options] A maximum line speed of 153 kb/s is supported. @b[All] lines on a switch may run at this speed. The interface options supported are: V24, V35 and X21. The interfaces are supported by using different Line Interface Cards and may be freely interchanged. In order to allow the V35 option to fit on the same size of card as the V24 option, a non-standard presentation is provided using a V24 style socket. The switch supports both internal and external timing, with internal clock rates of; 2.4, 4.8, 9.6, 19.2, 38.4, 76.8. @section[Throughput] A throughput of 581 data packets through the switch is claimed, this performance was observed while using a traffic generator by Telefile itself. The switch loaned to the ERCC achieved 400 data packets/sec, limited by the nature of the test equipment and limit of 8 lines. @section[Costs] @begin[verbatim] Model No. No of Lines Cost Comments --------- ----------- ---- -------- T3639-001S 10 @T{#}8,450 This unit has only 128K of store and is not expandable. T3639-101 8 @T{#}14,900 2 Mb of store expandable to 32 lines. T3639-102 12 @T{#}17,000 - do - T3639-103 16 @T{#}19,300 - do - @end[verbatim] In addition to the above, if a floppy disc is required, it will cost @T{#}1,860. A two link extension costs @T{#}1,300 (including interface modules) Hence, for the standard configuration: @begin[verbatim] 10 lines @T{#}8,450 18 lines @T{#}20,000 48 lines @T{#}50,400 @end[verbatim] It is noted that the 48 line system requires two boxes. @section[X.25 Facilities] Fast Select, Window size negotiation, PVCs and Reverse Charging are supported. Call Statistics are not currently supported but are targetted for the January release. Call Redirection is also targetted for the January release. Packet size negotiation is performed but a call requesting a packet size of 256 to be output by the Telepac, will be negotiated down to 128 bytes by the switch. When a packet containing more than 128 bytes is received by the Telepac it is repacked into two packets (using the M bit). This is a limitation caused by the current memory boards and will be removed when the 2 Mbyte boards are supplied. Full load sharing is planned for the January 1986 release. @section[Multinode Networks] Multinode networks can be supported using 'simple' X.25 to interconnect the switches. Any line can be a DTE or a DCE. It has been pointed out to Telefile that using this strategy means that Network Clearing codes are lost and they are investigating a further configuration option that would allow a connected DTE to pass on Network Clearing codes. @section[Addressing] An addressing table is held in the battery backed RAM and can be amended at any time without the need to reload the switch or even restart a line. The form of the addressing table is a port name (actually the name of a logical channel - but in a simple configuration there need be no difference) and an X.25 address. If less than 12 digits are supplied as the address then a match is performed on the digits supplied. This allows ranges to be entered - but on a restricted form, ie. the range 7250 to 7275 could not supported, only 7200 to 7299. Multiple entries for a single 'line' may be entered. It has been pointed out to Telefile that this scheme means that 1234 and 000000001234 would be different addresses and hence need multiple entries in the routing table. It is planned to strip off leading zeros in the January release of the software. @section[Network Management] Modules exist within the switch to collect various network management statistics and to direct this information to the Network Management centre. The information is passed over a 'simple' X.29 call, ie. all the information is passed in a purely text form that could be dumped to a printer. In the current software this call MUST be made from outside the switch into the management module, thus preventing the use of a printer simply attached to a PAD. Telefile will offer the first release of a network management package, in January, designed to run on an IBM PC with the X.25 card produced by Forge. The cost of this software will be @T{#} 1,000 per site. This software will also poll the switches for performance figures and error conditions. As with the statistics, this is achieved by transmitting the command in text as would be issued by the operator at the control console and analysing the text response from the switch. Between 30 and 40 nodes could be controlled from this terminal, or if desired, the software could be implemented in a user's host machine. @section[Statistics] The current software only counts the number of calls, data packets and CRC errors but Telefile have stated that the rest will be added in the January release of the software. The switch complies with the rest of the points in the specification. @section[Operator Control] It is not yet possible to disable a link, although it has been stated that this is planned for January 1986. The status of a link can be determined. It is possible to clear down a particular logical channel although the ability to display addresses in use is scheduled for the January release. Network addresses can be changed without restarting the line whose address is being altered. The change of line speed requires the line interface card to be removed and plugs changed but this can be done on a live system. The current software does not have password control on remote access to the control functions of a switch although Telefile have promised to add this feature in the January release. One restriction is that only one 'operator' may be active at any one time. @section[Date-Time STamp] Messages are date and time stamped. The date and time is held in battery backed RAM Compatibility, and is not lost when the power supply fails or the system crashes. @section[PSS Capability] Telefile have stated that they are prepared to guarantee compatibility with PSS. @section[Performance - Packet Transmission] @begin[verbatim] Call Requests 47.0ms Data Packets 6.2ms Clear Request 16.8ms @end[verbatim] @section[Availability and Reliability] It is estimated that the mean time between failure for hardware reasons is about 6000 hours. Telefile will be prepared to guarantee a MTBF of 2000 hours for all reasons. The test switch in ERCC has run for almost 3 months with no failures. @section[Operation] The system is currently loaded from EPROM and the configuration from battery backed RAM. No dumping is currently performed but this will be available on systems that include a floppy disc. @section[Reconfiguration] The reconfiguration tools are extremely flexible, although the current documentation needs upgrading in order to make it more obvious which of the available options should be used to create a switch that performs as expected in the Community environment. @section[Known Problems] There a number of points of variance with the JNT spec but Telefile have agreed to change their software to match the requirements by the January release of the software. A number of problems have already been resolved since a switch was loaned to the ERCC. @section[X.25 (1984)] Telefile are prepared to upgrade the switch to ISO DP8878 when requested by a customer. This could be made a requirement of the first few procurements. @section[Academic Community Special Requirements] By early 1986, it will be possible to link up Telefile switches by parallel channel interfaces. One potentially useful option would be to mount the conversion aids in a separate M68000 processor linked in such a fashion. @section[Product History] The current switch has been on the market for about 3 1/2 years but has not been marketed as an X.25 switch to any extent. Approximately 100 systems have been installed worldwide, including 30 in the USA and Canada. The Telepac software is also used by Motorola in their switch product. @section[Maintenance Arrangements] The cost of hardware maintenance will be 8% of the capital cost offering a 4 hour turnout period. This work would be done by SEEL in Scotland, a third party would be 'signed up' in England. A high quality of service, with a 2 hour turnout, costing 12%, would be offered in some locations. A 'repair and return' service would also be offered. The software will be maintained free of charge for an agreed period of time - five years was not considered unreasonable by SEEL but this will obviously need to be tied up in an agreement. @section[Company Information] The UK company has 30 employees, six of whom work on developments. The US company, whose origins lie in providing look-a-like computers to Xerox mainframes, takes no part in the switch developments. @section[Contact Name] @flushleft[R. Chisholm, SEEL Ltd., 3 Young Square, Brucefield Industrial Park, Livingstone, EH54 9BJ. Tel. No. 0506 411503.] @section[Confidentiality] None. @section[Summary] Although the switch does not yet fully meet the JNT spec (Dec.'85), it meets more of it than the alternative small switch suppliers. Although the price is. higher than CAMTEC it is lower than Dynatech and GEC and represents good. value for money.. A number of features have been promised in the January release of the software. and it will be necessary to check that these have been implemented. @set[section=0] @newpage @majorheading[Telematics] @section[General] Telematics offer a range of X.25 networking products called the Net 25 Product Family including concentrator switches, primary switches, backbone switches, PADs and supervisory nodes. The equipment is sold in the UK by Telematics and also by ICL and other companies including IAL who sell it under their own label. Telematics is certified for use on most public data networks. @section[Hardware Description] The Net 25 switches are based on a proprietary 32 bit bus that is stated to run at 32 Megabytes/sec. There are three models available, the P500, P1000 and P2000. The main difference being the number of processors used and the number of slots available in the back plane. The particular switch configurations, described as Net 25/XXX are constructed from the appropriate model. The Net 25/100 switch consists of the bus and a general purpose M68000 processor card, a memory controller and memory, either a floppy or a hard disk, and a Network Communications Processor (NCP). A Net 25/200 is derived from a P1000 and includes a 'communications accelerator' which is a two part cache, one part a cache for the general processor and the other part a cache for the NCP. The Net 25/400 is formed by the further addition of one or two additional general processors. The Net 25/800 uses the P200 with an additional two processors. The series are field upgradable to the P2000. The P500 has 2 Mbytes, the P1000 8 Mbytes and the P2000 16 Mbytes of store. The NCP supports a TDM based bus with a scanner with a maximum capacity of 2 M bits/sec. The scanner can support up to 60 'Line Interface Modules', each supporting up to 8 lines (depending on speed). In addition a new module, called a Line Programmable Module (LPM), has been delivered to some customers, although there is no BT approval as yet, with 120 Kbytes of on board RAM. In addition to the support of 2 V24 or V35 ports, the LPM supports up to two expander cards, Line Programmable Extenders (LPE), to which either 8 V24 or 4 V3 lines can be connected. In all cases the level 2 handling is done in one of the 68000 processors. @section[Software Description] The basic operating system, called TRAX, is used to support a number of tasks which implement the switch software. In addition it is possible to run different complexities of Network Management tasks or 'user' written tasks. (A 'user' task is any not written by Telematics.) TRAX is written in M68000 assembler code. The switch software, consist of a number of tasks to handle call set up, handle data transfer and network management an Interface 6 Network Management (INF). The data transfer task is written in assembler and the call set up task is written in C. 'User' tasks can be written in either C, Pascal or assembler. @section[Configuration Options] The limits for the number of line cards per switch are as follows:- @begin[verbatim] Line Cards V24 Line V35/X.21 Lines ---------- -------- -------------- P500 (Net 25/100) 8 64 32 P1000 (Net 25/200) 60 480 240 P1000 (Net 25/400) 60 480 240 P2000 (Net 25/800) 60 480 240 @end[verbatim] The figures are for the number of lines assuming all the lines are of that type, a mixture may be created by assuming 2 V24 lines for every V35 ones. These configurations assume the use of Line Interface Modules (LIMs) rather than Line Processor modules (LPMs) and Line Processor Extenders (LPEs). If a configuration uses LPMs and LPEs then a 'Line card' can be either LPM supporting 2 V24 or V35 lines or an LPE supporting 8 V24 or 4 V35 lines, bearing in mind that a single LPM can support either one or two LPEs. Thus the consequences of using each LPM is to reduce the maximum connectivity of the switch by 6 V24 lines or 2 V35 lines. The advantage of configuring a switch with LPMs and LPEs is that the constraint of the TDM nature of the NCP is avoided. @section[Line Speeds/Interface Options] Telematics support 64 kb/s in addition to the line speeds in the specification. V24 is used up to 19.2 kb/s and V35 above that. In addition X.21 (in reality RS449) is supported. Both internal and external clocking can be used at any of these speeds. The USA company is doing a T1 (1.56 Mb/s) and the UK company have 'requested' that a G703 interface is manufactured but they consider that one of the UK OEMs may do it first. @section[Throughput] The following throughput is claimed for the Telematics range. The figures are for data packets only through the switch. There is no dependence on packet size. @begin[verbatim] Net 25/100 100 dpp/s Net 25/200 200 dpp/s Net 25/400 400 dpp/s Net 25/800 800 dpp/s @end[verbatim] Figures are available for the Net 25/200, based on the switches for BT, which show a packet throughput of 280 data packets/sec. @section[Costs] The following List Prices have been quoted by Telematics for the 4 configurations. @begin[verbatim] 10 line: @T{#}30,000 (actually 14 links) 18 line: @T{#}31,000 (actually 20 links) 48 line: @T{#}86,000 (actually 50 links) 100 line: @T{#}135,000 @end[verbatim] This cost assumes a 'C' level of software licence, there is no direct support with this level so there is an implication that at least one university would buy the full support - the 'A' level - and co-ordinate the bug reporting for the rest. The alternative is that at least one 'A' (but only one) licence must be bought by each university. The cost of the switches with an 'A' licence are as follows: @begin[verbatim] 18 @T{#}41,406 48 @T{#}96,406 100 @T{#}145,406 @end[verbatim] The costs include software with the 'lower' level of network management (single switch only). A single licence for the full network management costs @T{#}9,307 and @T{#}23,000 for the 'C' and 'A' licences. The hardware quoted for the 10 line switch is about to be replaced by a new option and reference should be made to ICL for details of it. In addition to the above, various items can be added to the configuration - @begin[verbatim] Extra processor (all) @T{#}10,500 Additional line cards (all) LPM @T{#}5,500 LPE(V24) @T{#}2,500 (V35) @T{#}3,000 @end[verbatim] Any one of the configurations can be field upgraded to larger configurations. @section[X.25 Facilities] Fast select, PVCs and Reverse Charging are supported. Packet and Window size negotiation is supported with a maximum packet size of 4096. In addition the network will segment or reconstruct packets, using the 'M' bit, to allow, for example, a host with a packet size of 4096 to communicate with a host using a packet size of 128. Call redirection is planned for 1986 as part of the X.25 (1984) enhancements. Call Statistics are not compatible with those on PSS. @section[Multinode Networks] Telematic switches connect together to make up a multinode network using a non X.25 protocol between the nodes. Adaptive routing is supported and if an alternative path exists, existing calls are not lost, or reset, when a 'transit' node, or link, goes down. Any node can control the rest, see the section on control. @section[Addressing] Multiple entries are allowed in the address table. The table size is only limited by the overall store size. An individual entry can be either a single address or can be made to match a 'range', eg. 1234 would match any address from 12340000000 to 12349999999. This is not quite as flexible as a 'range of addresses' but by using multiple entries, the differences are fairly minor. A problem is caused because the addresses 1234 and 000000001234 are different. To overcome this will require double entries for all routing addresses. Full load sharing is not supported. @section[Network Management] Network Management can be performed on a separate machine or the software can be integrated with the switching software on one of the switches in a network. If it is integrated then there could be a performance cost although it is possible to run the management package either at a priority level less than that of the switching functions or run the software on a separate processor within the switch system. No explicit charging is done by the current software though the information is present to do charging for calls, for example, to PSS. The management protocols are @u[not] in the public domain, although Telematics would sell us listings etc., to allow community hosts to perform some of the management. It appears that they would @u[not] allow other switch manufacturers to implement the protocols. @section[Statistics] Telematics meet the JNT spec except that there is no explicit code to calculate the percentage utilisation of the switch as a whole. Telematics do not report in real-time the number of active calls on a link or the respective addresses. @section[Operator Control] It is @u[not] possible to clear a call to a selected address, the rest of the spec is met with the ability to change network addresses on a line link. It is possible to use any network terminal (password protected) for switch control. There are two levels of Network Management/Control Software - The Remote Node Control and the full Network Control Centre code. The first is a partial form that only allows the operator of a particular switch to control that switch. The second adds the ability to look at and control remote switches. At lease one node in each university with multiple nodes, would require either one copy of the Network Control Centre software or a separate support management machine. @section[Date-Time Stamp] Complies. @section[PSS Compatibility] Telematics say that they are currently compatible with PSS. However, if in the future there is a divergence between PSS and CCITT then Telematics will @b{guarantee} compatibility only with CCITT. In reality, so long as a difference causes problems with more than just the Academic Community, Telematics will fix it. A difference between PSS and CCITT which only affects the Academic Community will not necessarily be fixed by Telematics. It is suggested by Telematics that purchasing through an OEM, such as ICL, could provide a resolution to such problems. @section[Performance - Packet Transmissions] @begin[verbatim] Call Request 30ms Data Packet 4-6ms Clear <30ms @end[verbatim] @section[Availability and Reliability] It was claimed that the Net 25 series would meet the reliability section of the specification. It was noted that line cards could be replaced on live systems, so that only a failure in a central module would force an entire node to go down. Redundant systems, ie line cards shared between two nodes could be constructed if necessary. Only figures for hardware reliability are stated - in terms of 7000 hours MTBF for the kernel system. It is claimed that field information to date backs up that figure. An availability for the largest system, (excluding line cards as they can be replaced) is calculated , and would be contracted on, at an 99.97% (hardware only). This figure assumes an MTTR of half an hour - ie., user holds and changes cards. Statistics on software MTBF will follow, but it is claimed that entire failures are extremely rare. entire node is quoted at 99.9%. @section[Operations] It is not possible with the current software automatically to dump the store to backing disk in the case of a complete failure. If failures of individual tests occur details of the failure, including partial dumps, are printed out on the system console. There is not a separate handler for each link in a system and consequently it is not possible for an individual link to fail, and require a reload, as occurs on the GEC switches. @section[Configuration] All configurations, including the initial set up, are done by using the INF software. This can be achieved by either inputting commands that directly change the configuration or by editing a file and then 'executing' the file. In the case of the file, only the differences between the file and the running system are actioned. @section[Known Problems] @begin[itemize] Management protocols are not in the Public Domain. Guarantee conformity with CCITT, not PSS. Not possible to list or clear calls to a particular DTE address. No dump procedure. No Full Load Sharing. @end[itemize] @section[X.25 (1984)] Work is currently in hand to implement X25 (1984). Features will start to appear in early 1986 and there will be a comprehensive implementation by the end of 1986. The OSI CONS document has not been studied. @section[Academic Community Special Requirements] The TRAX system together with the X25 system has been designed and built with the specific criteria that it is easy to add 'user' modules to the system to tailor it for non standard requirements. However, Telematics state that they will not have enough spare effort and would not wish to take on 'specials' for the academic community. They stated that various software houses could do specials but they were not prepared to take prime contractor status for the work. The advantages and disadvantages of the community 'doing it itself' would need to be carefully weighed up. @section[Product History] The Switch first appeared on the market at the end of 1984. There are now 250 to 300 systems installed world wide. The UK customers include BT with a 90 system in VASSCOM, designed to carry viewdata traffic. The major USA customers include Telecom General and Geisco. @Section[Maintenance Arrangements] The cost of hardware maintenance is approximately 7% of capital cost, the following figures were quoted for the standard configurations:- @begin[verbatim] 10 line @T{#}2,040 annually 18 line @T{#}2,112 " 48 line @T{#}5,940 " 100 line @T{#}9360 " @end[verbatim] If the 'C' level software licence is used, there is no cost for software maintenance. This assumes that one university in the community would hold an 'A' level licence and act as a co-ordinator for fault reporting from other universities. If the 'A' level of software is used then there would be an additional cost of @T{#}3,948 annually per university (regardless of size of switch). It is possible for the user to do 'first line' maintenance. @section[Company Information] The British Company is a wholly owned subsidiary of Telematic International Inc based in Fort Lauderdale, USA. The USA company designs and builds the hardware, although it is planned to manufacture in the UK during 1986. The British Company does the X.25 software development. The current UK company size is 30 - 35 people with a team of 9 working on X.25. @section[Confidentiality] Details of higher speed link plans. @section[Contact Name] @flushleft[D.N. Birss, Telematics International Ltd., Intec, Wade Road, Basingstoke, Hampshire, RG24 0NE. Tel No. 0256 567385] @section[Summary] Telematics offer a very credible alternative to our current switch suppliers. However, there are still a number of questions to be resolved, by further discussions, including the decision as to whether they could be bought through a third party, for example ICL, who could offer enhanced facilities and different license and support costs and arrangements.