Systems Network Architecture (SNA)
Introduction
IBM mainframes were originally made with fast-coupled devices to obtain the necessary speeds. Nowadays CMOS circuits
are fast enough to be used. The operating systems used are commonly MVS, DOS and VM, 95% of users use MVS. Mainframes
from IBM, Amdahl and Hitachi are all plug compatible, bar a few minor differences. For communications, MVS uses
VTAM which stands for Virtual Telecommunications Access Method and is the software that runs the network. IBM do not use ASCII, only EBCDIC.
- The Channel Interface is extremely powerful. It is a 64 (or even 128) channel processor that can handle 5 disk controllers
at a time (each disk controller could have 10 disks).
- The main task of the Front End Processor (FEP) is queuing, it runs NCP.
- The Bus and Tag connection is a copper connection, normally thick grey coaxial cable bundles. Escon is 'fibre-channel'
and is capable of carrying up to 17Mbytes/sec.
- Cluster Controllers, or 3174s, can run straight from the main processor, or from the FEP. The screens are commonly 3287s
which buffer all input and send it one go.
- Controllers and terminals are Physical Units (PU) and there are a number of Logical Units (LU) for each PU. The 3174
is PU 2.1 and the screens are LU 6.2. There are also LUs 2, 3, 4, 5 etc.
CMOS mainframes can be segmented into Logical Partitions (LPARS). The partitioning is measured as a value of MIPS and is
effectively a way of dividing up the processing power of the main processor. Different disks can be assigned to different
processes via the Escon Multiple Image (EMIF), because normally Bus and Tag will only allow connection to one
machine.
Sysplex is a new way of connecting mainframes.
SNA Operation
Systems Network Architecture (SNA) was designed by IBM to include everything from the VTAM to the LU.
LUs are synonomous with TCP/UDP ports in the IP environment. Applications such as CICS can report errors or failures. The guaranteed delivery means that no
packet is lost, or if there is a problem, it can be traced. This makes it very attractive in the banking world.
An Initial Program Load (IPL) is equivalent to a boot. The VTAM has to be told about all the devices attached before it starts.
This achieved by performing a Gen. This is where all devices are entered into a text, or source, file and this is assembled
by the VTAM by an assembler. When someone wishes to be added, they have to wait until the next time a gen is done, typically
once a month on a Sunday night, often a pressurised time. The connection structure is hierarchical.
When PCs arrived on the scene Token Ring was developed for the 3174 to save having to define individual PCs on the 3174
in a gen, only the token ring itself need be defined. Eventually, this progressed to being included on the FEP, along
with other interfaces such as X.25, albeit reluctantly by IBM.
In an NCP gen, the NCP source file is assembled by the VTAM. The text file contains devices with their naming conventions,
part numbers etc. Copy and paste could sort of be done from previous files using cc.
In addition, window sizes, sliding windows, routing tables (both outward bound routes and inward ones), segmentation etc.
are all statically built.
After an IPL has occurred banner messages are sent to every screen, so if there are 2000 devices the network slows down
such that no screen will operate for perhaps up to 30 minutes.
You can have FEPs arranged in triangulations, you could have 50 FEPs all interconnected, this consequently makes the
routing tables extremely complex and tedious to write. There are whole programs that are very expensive written just
to deal with writing routing tables. Techniques such as sub-area routing have been introduced, a FEP could be sub-area.
Say you needed to have a supermarket and a bank interconnected such that the supermarket could see their bank details immediately. For
this, two domains could be created, one for the supermarket and one for the bank, and these domains are then connected with a serial link
called a SNI link.
Advanced Peer to Peer Networking (APPN)
PCs often run applications apart from the mainframe:
Advanced Peer to Peer Networking (APPN) was developed for this purpose. It allows a session to occur without having to
go through the hierarchical structure. APPN was too much, too late, it is very complex to set up.
Tunnelling SNA
Routers replace the FEPs and are normally LAN connected to the mainframe since a serial link would be a bottleneck.
Cisco started using Serial Tunnelling (STUN) as the bridging protocol for SNA, then moved on to using RSRB (Remote Source Route
Bridging). More recently this has been replaced by Data Link Switching (DLSw/DLSw+).
The advantage of DLSw is that it can:
- Spoof: Mainframes can be set to poll devices 10 or 100 times a second. DLSw acknowledges spoofing locally and
caches the explorer so that it does not go out across the expensive WAN.
- Do Netbios filtering:
- Has Expanded RIF and therefore hop count (RSRB does not scale!):
Cisco CIP/CPA
On the Cisco router, the Channel Interface Port (CIP for the 7500 or CPA for the 7200) contains a virtual Token Ring
which you define in the IOS. This port can handle 16,000 LUs/sec.
Using Switch Major Nodes on the main processor needs no special parameters, dynamic configuration occurs allowing
the VTAM to think that a remote PC is on the virtual LAN for instance.
On the router at the remote end some configuration is necessary when switching the 3174 serial connection across
from the FEP to the router serial interface.
This SDLC conversion is done in the IOS of the router remotely:
- sdlc
- C1xxx - device identifier.
- nrz
- MAC address of the virtual LAN at the other end. This will be set to something like 7500.abcd.1234 when you
configure the CIP interface on the router at the other end.
- XID (transmission identifier)
- ID Block
- ID Num
The last two items are assigned by the mainframe operators.
Gateways are there to convert ASCII to EBCDIC. Mainframes have to evolve to cope with an increase in the number of users, gateways,
IP data, multiple networks and the connectivity of PC emulators/terminals.
The computer is now the network!
DLSw needs to have a higher priority over other protocols. There is nothing worse than having differing responses from an individual's
keystrokes. Cisco to Cisco is imperative for Quality of Service to be delivered.
The CPA (for the 7200) is a half-size board compared with the CIP (7500). It has 2/3 of the throughput, say 7,000 concurrent users,
or 400 transactions/second.
In the Cisco IOS for the CPA (or CIP) you type claw1 (for IP connectivity to say Production); claw2 for IP connectivity
to Test and claw3 for connectivity to Y2K.
IBM3172 replaces Interlink boxes.
The CSNA is the virtual LAN in the card. You need to define Switch Major Nodes in the VTAM in order to recognise
PUs in the 3174s. Any SNA connectivity requires CSNA, and this requires a separate licence to be bought from Cisco.
12% efficiency is lost on encapsulation of the SNA, however this is made up for by header compression.
TN3270 is the screen emulation server and the PU can be in the CIP/CPA rather than on the network. The PU could also
be held in software in the IBM MVS TCP/IP if you wished, although this is not as efficient, also the Microsoft
SNA server could have the PU, but this keeps crashing.
32Mb of on board memory is sufficient to support 2000 concurrent users. 16,000 concurrent users would require 128Mb.
Notes
DLSw1 and DLSw2 are now RFC standards.
IBM have lost out on the peripheral devices over the years.
Token Ring cannot support IP multicast so is heading for extinction!
SDLC is SNAs equivalent to PPP (HDLC).
What about a 1000 customers who wish to look at their accounts overnight, even an IBM 2216 channel connector does not
have enough capacity.
TN3270E supports printers.
XCA Major Node in the VTAM sets up the virtual LAN between the CSNA and the VTAM.
|
