Implications of the Switch to X25 ================================= a) Hardware 1. TCPs and Workstations The following TCPs will need a store upgrade (amount in brackets) Bush 11/10 (32kb) Cs-at1 11/10 (16kb) Eng1 11/10 ( 8kb) G. Sq. 11/10 (16kb) Pollock11/04 (32kb) Scri-P 11/10 (32kb) The following TCPs will need an HDLC line Interface card Art In. 11/02 (dpv11) Cs-at2 11/23 (dpv11) E Lib 11/23 (dpv11) Eng1 11/10 (dup11) S. Fac 11/40 (dup11) WSCoa 11/10 (dup11) The Science Faculty TCP (S. Fac) will not be able to support the current number of users, the line printer & plotter on a DUP11 as on the parrallel interface so an extra TCP will need to be purshased. The slow devices workstation (Sdevs) will probably not be able to support as many devices (pdp11 addressability) and a device will have to be moved elsewhere. The SCRI-I workstation with its 8 users and RJE devices will also be short of store an additional problem is that this machine does not belong to us. A number of TCPs are currently down-line-loaded from a switch. It will not be possible to do this under X25 unless a special machine is provided that the TCP is manually switched to, the down-line-load performed and the TCP switched back. The alternative is to provide about 4 TU58 cassette readers for the non-ring TCPs. 2. Front End Processors for Emas The 2988 Feps are only 128kb, they will need to be upgraded by an extra 32kb in order to fit in the X25 software. Any increased load on the 2988's will have to be handled by upgrading one or two of the PDP 11/34s to 11/44s. There will be a problem with unbalanced load if Glasgow University still require a NSI FEP beyond October '83. This would mean that the one FEP running NSI would be underutilised with a corresponding extra load on the other two FEPs which they are unlikely to be able to take. Both of the PDP 11/44 2972 Feps will require a DQS11 pair of synchronous line handlers but these should become available as the existing switches are dismantled. 3. Other network Hosts and Terminals Edinburgh Filestore This machine is being moved to the Ring and will not require any further changes. PSS Gateway - Already on the Ring. XGate (Ring to X25 gateway) - This machine will require a DQS11 synchronous line unit as the load increases (from existing switches). ERTE - Software changes only. Unix Hosts (EEVAX & EMCU) These hosts will either require a complete network software rewrite to X25 or the X25 11/03 FEP should be bought from York. Prime Machines (Abro & Arcus) - These can both be easily switched. 1130 Packages Stirling University: Under discussion Physics 11/45: The future of this machine is under consideration, if it is switched to running RSX11 then suitable packages can be purchased. Key to Disc: MDS are working on a X25/X29 package. An alternate solution would be a 'black box'. Oban (dial in): Under discussion East of Scotland College of Agriculture: They have applied for a Vax. SIAE (bush): This is an RSX11 system, so a package can be bought from DEC if there is sufficient store. INFO: Info will require a fairly major rewrite. 4. Packet Switches Currently there are two PDP 11/45 switches both sited at Kings Buildings (KB), their maximum packing switching capacity is approximately 270 packets/sec although it is not desirable to run them at this figure as it would lead to excessive queuing for the CPU. One of the existing limiting factors is the bandwidth between George Sq. (GS) and KB. The current provision is two 48kbaud links multiplexed down into predominantly 9.6Kb circuits. The demand has been growing in the GS area and there is only very limited opportunity for splitting the circuits into 4.8Kb ones. The only feasible solution is to site a switch in the KB area with a second switch sited at GS. Sizing The peak load measured on the existing switches is 210 packets/sec averaged over a five minute period, the instantaneous peak is obviously higher. At this load both of the switches are seriously overloaded and are in themselves imposing delays in user response. In addition to this there are a number of factors that necessitate newer switches with a far higher throughput. 1. There are a number of new machines that are, or are just about to be, connected to the network, for example the ERCC Vax and the Medical Computing 11/44 running Unix. 2. There is a lot of pressure from users to have their terminal speeds increased from the existing 1200 baud to 2400 baud or 9600 baud which will generate extra traffic. 3. The amount of screen based work done across the network is limited, mainly because of poor response. With more network capacity available the number of screen based programs, including scrren editors, will increase and a large increase in network traffic will follow. 4. The number of interactive users on the 2972 is currently effectively limited by poor comms response. The diagram shows the estimated load in October 1983, it uses two assumptions, firstly, there is no completely 'new' load and secondly, measured figures are used wherever possible so that there will be an increase as the switch bottleneck is removed. On this basis the minimum throughput for the KB switch is 192 packts/sec and 136 packets/sec for the GS switch. These figures, of course, are a base load, the peak figures would be considerably greater. If the switches are to be able to support the load for any length of time, and in particular if they are supporting the major load for a new ICL mainframe with say 240 users then it is reasonable that at least double or treble those figures are achievable on the new switches. A total of 32 lines (including the new Memex machine & a link to Strathclyde) are required on the KB switch, 6 of these lines would be at 19.2Kb and 26 at 9.6Kb or less. The GS switch requires 18 lines, 4 at 19.2Kb and 14 at 9.6Kb or less. It would seem prudent to install at least 38 on the KB switch and 24 on the GS switch to allow for an spare communications card on each switch for resilience. Software Implications 1. RJE The existing RJE protocol consists of two distinct parts, the protocol used to transfer files from one machine to another and the user interface that is used to associate logical queues with physical devices and control the flow of files to those devices. It is proposed that FTP will be used to transfer files if it is physically possible to fit an implementation with the PDP11 workstations. The user interface would be implemented using the standard interactive protocol to log onto the host machine. 2. ITP A switch to XXX from RCO-ITP would require a number of significant changes in the appearance of the service to the user, if, in addition, JNT pads were used there would also be a problem with network accounting. A more detailed description of the differences are included in the appendix. A particular feature of ITP is the multiplexing of a number of user conversations into one virtual call. If XXX were used it is unlikely that the EMAS FEPs would be able to support about 80 simultaneous calls each on the current hardware without a complete rewrite of the software. For these reasons it is proposed that ITP is used for the majority of the users on the X25 network although all the EMAS FEPs will also support XXX calls. Any user accessing a purely XXX host, such as the ERCC VAX will be gatewayed through XGATE.