TINE (Three-fold Integrated Networking Environment)
- (TINE++ % 4) = INET and Remember: This Is Not Epics!
But you can run EPICS iocs on TINE using Epics2Tine.
TINE is embedded in DOOCS, so you can also run DOOCS clients and servers using TINE.
TINE can also be used in a STARS system and via a STARS-bridge in a COACK system.
You can also include TANGO elements on your TINE system using Tango2Tine.
But you might want to go native ...
Current Release level: 4.6.1
TINE web client applications can be easily configured and run on any browser using Web2C
TINE is fully supported by ACOP (click here for java doc), Abeans, JoiMint, and Control System Studio.
You may want to have a look at the release notes for version 4.00 or take a quick look at a Bird's Eye View of TINE.
Download TINE here.
Questions or comments can be addressed to firstname.lastname@example.org
Bug Reports can be submitted to the tine tracker.
General discussion or questions can be submitted to the tine forum.
Note: Due to an ever increasing amount of 'forum spamming' you will need to send a separate email to email@example.com requesting a forum account if you wish to actively participate.
Perhaps the most distinguishing feature about TINE is its integration of client and server components of vastly different networking environments. To begin with, TINE is a multi-platform
system, running on MS-DOS, Win16 (Windows 3.X), Win32 (Windows 95,98, NT, 2K, XP), UNIX (Solaris, HP-UX, OSF, SGI, Linux, FreeBSD), MACOS, VAX and ALPHA VMS, VxWorks, and NIOS. TINE is also a multi-protocol
system to the extent that UDP, TCP, IPX, and PIPEs are all supported as data exchange. Finally TINE is a multi-control system architecture
system, allowing client-server, publisher-subscriber, and producer-consumer data exchange in any variation. We shall describer these in more detail below.
TINE runs on a number of platforms, and can be thought of as a 'software bus', meaning that one can intermingle platforms at will. In general the members of a client-server pair have no knowledge as to the platform of its partner. Note that for small systems or subsystems it usually makes more sense to stick to one specific platform for front-end components and/or console components. Although there is nothing that precludes using a heterogeneous mixture, issues of maintenance (where a small number of persons are responsible for a large number of components) come into play. At HERA, the de-facto console platform is Windows XP (Window 3.1 in earlier years). All manner of front-end platforms are in use, but the individual sub-system laboratories (e.g. the RF group) stick to their preferred platform for all sub-system components. There is sometimes a point to using the same platform for console as for front end. In such cases, the console application can then run directly on the front end to aid in system diagnostics and trouble shooting. Where this is not the case (either not possible or not practical), one needs to either have an available console in the general vicinity of the front end, or separate diagnostic tools available on the front-end platform.
Note also that by allowing a heterogeneous system, expensive front-end hardware can be used where warranted (for mission-critical devices) and inexpensive hardware used elsewhere. Furthermore a systematic, piecemeal upgrade of a control system is possible, since TINE will run fine on older systems such as VAX-VMS and MSDOS as well as the more modern systems such as Window XP, Linux, Solaris or VxWorks.
TINE supports both IP and IPX ethernet protocols. Generally, the IP protocol is available across all platforms, whereas IPX is available primarily on PC systems (DOS, Windows, Linux). Where necessary, IPX could also be ported to the other platforms. However, an IPX-stack must be obtained and installed separately, as it is not in the standard system kernels for these cases.
TINE defaults to UDP datagrams for data transfer. In most cases this is fine. If client-server communications occurs over a 'lossy' network, you may want to resort to TCP streams, at least for commands which change settings. This requires only an additional flag to be set in the communications API calls. The default data transfer can also be configured by the environment variable TINE_TRANSPORT (e.g. TINE_TRANSPORT=TCP). There are actually two kinds of TCP. Simply specifying "TCP" signifies a payload transfer which can be 'parceled' and reassembled (as per UDP) if the payload is large and which dutifully respects all 'timeout' parameters. By specifying "STREAM", a TCP/IP transport is used which passes the entire payload on to the local network stack and only times out at the connection establishment level (only available on multi-threaded builds). A local pipe or memory-mapped file is used if the transport is from a client-server pair on the same host machine.
Tine supports three modes of data exchange, each of which could be used individually to define the control system architecture. More likely, you will want to use these modes in combination.
- Client-server : A traditional data exchange mechanism available in most control systems is pure, synchronous client-server data exchange, where a client makes a request and waits for the completion of the request. This is not only traditional, but necessary when sending commands to a front end, where the next action to take depends on the outcome of the command. If however this is used as the sole basis for data exchange, the response on the client-side can suffer (dramatically) when a server goes down or network problems arise. In such a case, all communication directed toward a server will time out, and due to the synchronous nature of the communication, the end-user must wait for the timeout period to expire before say keyboard or mouse input can be issued to the client process. Furthermore, if several clients want the same information (regular updates of control data for instance), a server will see each request from each client separately. This can become a burden to the server if many clients (say 50 or more) are all getting a kilobyte's worth of data at 1 Hz, as the server will have to acquire the data at 50+ Hz.
- Publisher-Subscriber : For many cases, a much better approach is the publisher-subscriber data exchange. Here a client (the subscriber) communicates its request to a server (the publisher) and does not wait for a response. Instead it expects to receive a notification within the timeout period. This can be a single command, or for regular data acquisition it can be a request for data at periodic intervals or upon change of data contents. In this format, the server maintains a list of the clients it has and what they are interested in. Now if many clients all want the same kilobyte's worth of data at 1 Hz, the server must acquire this data set only once per second, and notify the clients on its list. This is much more efficient than the client-server model in such circumstances.
- Producer-Consumer : A third alternative for data exchange is the Producer-Consumer model. In this case a server is the producer. It transmits its data via broadcast on the control system network or via multicast to a multicast group. Clients (i.e. consumers) simply listen for the incoming data. This is frequently the appropriate data transfer mechanism. For most control systems, there are certain parameters which are of system-wide interest. At HERA for instance, the Electron and Proton beam-energies, beam-currents, beam-lifetimes, states etc. are made available via system broadcast at 1 Hz. Such read-only, machine parameters are of vital interest to a large number of running applications and front ends. If they were not available on the network in this manner, then some poor server would have to supply these data to 100 or more clients at 1 Hz. In "REFRESH" mode in the producer-consumer model this would by no means cripple the server, but would just be a waste of bandwidth on the net and CPU load on the server. In some cases, a good multicast is what is called for. Take note, however that if a broadcasts are used, then those clients which are to receive the broadcasts must reside on the network where they are found (the designated control network). If multicasts are to be used, then all routers should support multicasts. Those clients not on this network or not a part of the multicast group must obtain the data via other means. This mechanism can be thought of as a 'poor man's' read-only reflected memory.
- Producer-Subscriber : A hybrid between the above two modes is also possible under TINE, in which subscribers request data to be produced on the network (a "network subscription"). See the discussion of the CM_NETWORK (i.e. CM_BCAST or CM_MCAST) control mode bit below.
If the control system name server is up and running, TINE clients and servers participate in a plug-and-play scheme for address resolution.
A new server if properly configured can "plug" itself into the control system database maintained at the TINE name server, without required administrative intervention. At startup, the server name and all equipment module names are sent to the name server, along with the server's IP (and IPX) address, port offset, and other descriptive information. The name server will check its database for name and address collisions. If the name server sees that a server is trying to reuse an existing name, the name server will attempt to contact the existing entry. If this entry does indeed respond, the name server does not update its database, but instead sends an "address in use" message to the server starting up. On the other hand, if the previous entry does not respond, the name server assumes that the front-end server (or equipment module) is being moved to another location and allows the address change to be made.
When a TINE client first attempts to contact an equipment module, it sends the equipment module export name to the name server for address resolution. If the name server can identify the equipment module, it returns the address information. If not, the client then resorts to its local database. If an address is still not found, the error message "non existent element" is returned. If a match is made however, the address information is cached locally at the client. Subsequent attempts to contract the same equipment module obtain the address from the local cache.
If a link to a server goes down, the client will generate timeout notifications. After several consecutive failures, the client will again attempt to acquire the address information. In this way, a server process can actually be moved from one machine to another, without requiring a restart of the client.
A TINE server will look for the environment variable FEC_HOME
to establish the local database directory. (This supercedes the legacy variable FECDB). All server-specific .CSV database files should be located in this directory. If this environment variable is not set, then the server will look in its startup directory. TINE Clients look for relevant .CSV files according to the environment variable TINE_HOME
. We note here that TINE servers take on the behavior of clients at startup when they register their services with the equipment name server (ENS). Thus both settings are relevant to TINE servers. Furthermore if a server is keeping local histories according to specifications in the 'history.csv' configuration file (or via the AppendLocalHistory() API call), then the environment variable TINE_HISTORY_HOME
is used to determine the repository for the long term history data. If this variable is missing, and its legacy equivalent (HISTORY_HOME or HISTORYDB) is missing then the location specified by FEC_HOME
will be used. These environment settings should include the database path up to the final slash "/" (UNIX) or backslash "\\" (DOS, WINDOWS). Thus you might have
set FEC_HOME = ~/database/
in the UNIX world, or
set FEC_HOME = C:\DATABASE\
in the DOS, WINDOWS world.
The identity of a particular device is needless to say an important bit of information regarding data exchange between a client and server. A client application will typically make a request to a device via its full device name and not further concern itself with the location of the equipment module. However the devName argument entered in an API call (see ExecLink()
) needs to be resolved into a specific device being serviced by a specific equipment module running on a specific front end server.
As a case in point, a client might want the beam position, i.e. property "POSITION" from device "WL167" (to use the HERA naming convention). So devName might be specified as "BPM/WL167" to denote the targeted device on the "BPM" device server. In this case, the default context is assumed. The devName might also be specified in full as: "/HERA/BPM/WL167". In either case, the system kernel must be able to find the "BPM" device server, and determine that it is located on a local equipment module called "BPMEQM", which runs on a front-end computer called "BPMSRV". The system kernel finds the latter quantities by either consulting a name server or a local database. Thus, the BPM device server must be properly registered for this to work. That is, the information entered in the name server or database must match the information contained locally at the front-end server. If a request for the local equipment module "BPMEQM" comes to the "BPMSRV" front end and there is no such equipment module, then the error code "non existent element" will be returned.
To reiterate, the TINE name-space consists of four levels, consisting of context, device server, device name, and property. Both the context and device name are optional to the extent that the equipment name server will be able to resolve the address for the device server (if no context is given) if any entry matches the device server name specified. The device name (or absence thereof) is passed along the server and dealt with there.
When a server starts it will inform the equipment name server of its identity and its address, as if to say "If anyone asks for a device server named BPM under context HERA, tell him to ask for equipment module BPMEQM at my network address."
Thus, in establishing a front-end server in the control system, some thought must be given for the following quantities:
- FEC Name (also known as: Server Name) : the 16-character, system-wide unique name identifying the server, or rather the server process. "FEC" is an acronym meaning Front End Contoller. For platforms such as MS-DOS, Win16, or VxWorks this also translates into Front End Computer. The point is: on Unix, VMS, or Win32 there can be several independent server processes running on the same computer, each requiring a unique "FEC" name even though they reside on the same machine. Under most circumstances this name is not hard coded into a server, but is instead obtain through an initialization file (see fecid.csv or fec.xml). Any server can in turn contain many equipment modules. There are no restrictions on the FEC name other than it be unique (system wide) and not contain more than 16 characters. In the above example, "BPMSRV" was used.
- Context Name (also known as: Facility) : the 32-character, context specification. This supplements the Equipment Module Tag Name to determine uniqueness. If left unspecified, then the default context is assumed.
- Device Server Name (also known as: Equipment Module Export Name, Device Group) : the 32 character, system-wide unique name identifying the equipment module located on a particular server. This name is the actual point of contact between the client API and the equipment module in question. Under most circumstances this name is not hard coded into a server, but is instead obtain through an initialization file (see exports.csv). There are no restrictions on the Equipment Module Export name other than it be unique (system wide) and not contain more than 32 characters. In the above example, "BPM" was used.
- Equipment Module Name (also known as: Local Name) : the 6 character identifier tag serving as a point of contact among local routines on a given server. This name will perhaps require some explanation. In a purely object-oriented environment, this identifier would not be necessary, as various routines such as RegisterProperty(), SetAlarm(), ClearAlarm(), GetNumContracts(), etc. would all be methods of an equipment module class, and the reference to the equipment module in question would be clear. However, TINE is object-based and not object-oriented and the kernel on most platforms in written in straight C (not C++). The kernel must be able to de-reference the targetted equipment module (as there can be more than one module per server). Furthermore, as the (unique) Equipment Module tag name is not, in general, hard-coded at the server, it will not do as an identifier. The Equipment Module identifier is on the other hand hard-coded and as it is seen only locally, it does not require system-wide uniqueness. This identifier tag is restricted to 6 characters. This is largely historical and has the benefit that the underlying system headers need only contain this much smaller name (once an address is resolved) as opposed to both a context and device server name (64 characters). As the Equipment Module identifier can (and must) be hard coded on the server, it is entirely possible to run precisely the same server binary executable on different machines (distributed front end). In such cases, the startup files fecid.csv and exports.csv would have to be different on the different machines (so as to guarantee uniqueness). In the above example, "BPMEQM" was used. The distinction "Local Name" and "Export Name" makes sense when viewed from the perspective of the server. A server executable has a hard-coded "local" equipment function name that it wants to make available on the control system network. That is, it wants to "export" the equipment function. But the exported name must uniquely point to the local equipment function on the server in question. Note that in the above example, one could as easily have used "BPM" for the FEC Name, "BPM" for the Equipment Module Tag name, and "BPM" for the Equipment Module Identifier. This would work, but in the end might lead to confusion.
- Device Name (also known as location): the (up to) 64 character device name identifying the particular instance of the equipment module in question. This name is not resolved by the name server or local database, but is passed as is to the targeted equipment module. In many cases, it is desirable to work with device numbers rather than names, in such cases simple passing a string representation of the number is suggested (prefaced by '#').
The above discussion is illustrated in the figure below:
Finally, note that some naming schemes will want to consider device 'groups' rather than device servers. If a particular server is replicated N times in a given facility, so that there are N phyiscal servers (each with its own unique FEC name and exported device server name) it might nonetheless desireable to take a purely device view of this set of servers. In other words, server #1 might have the first 5 devices of a group, server #2, the next 5 devices, and do on. The TINE naming hierachy will let you define a device 'group' which acts as a logical server for this collection of individual servers. This uses the TINE redirection features and requires a running Group Equipment Name Servers (GENS), which can either be configured by an administrator or which allows plug-and-play requests from the booting servers. In the plug-and-play scenario, an initializing server will signal it's intentions to 'join' a device group.
So a client might make calls to power supply controllers on what looks like a device 'server' called "MAGNETS" and really be redirected to various phyisical servers ("MAGFEC1", "MAGFEC2", ... for instance) which really handle the requested device.