casey@gauss.llnl.gov (Casey Leedom) (12/08/89)
I just finished a massive rewrite of my remote X11 standards proposal. I got several requests for copies, so I'm going to post it here. Please excuse the length. Major changes are mostly in the area of providing more detail and making the proposed protocol much more specific to X11. ----- Thu Dec 7, 1989 ``XRPCP'' X11 Window System Server DIX/DDX interface Remote Procedure Call Protocol DRAFT Standard Proposal Abstract: This is a proposal for a simple, bandwidth efficient Remote Procedure Call Protocol, XRPCP, to implement the X11 Window System Server DIX/DDX interface. This is the interface between the device independent and device dependent code in the X11 server. This protocol would be used to facilitate the splitting of the DIX and DDX sections of the X11 Server into two cooperating processes. The primary motivation for this protocol is to allow Workstations to run X11 when they are remotely connected to Network Hosts via low bandwidth links and/or the Workstations don't support the software necessary to run X11. This would be accomplished by separating the typical X11 Server which directly manages Workstation resources into two parts: an X11 DIX Server, Xdix (or perhaps Xrpc?), and an X11 DDX Server, DX``platform''. Xdix would contain the X11 Window System Server/Client communications modules, high level graphics facilities, and graphics database. An X11 DX Server would providing remote access to the Workstation's resources (input handling, rasterization, etc.) An example would be an IBM PC with VGA to a Network Host via an EIA232 serial link. The personal computer would run, say, DXvga and the Network Host would run Xdix. XRPCP would of course not be limited to this application since it would provide general purpose access to a Workstation's facilities. However, this document focuses almost exclusively on this application in order maintain a concrete focus on our primary goal. Additionally, we will often use the TCP/IP network protocol suite in our examples for much the same reason. Whenever possible we will try to pull back to consider XRPCP from a more general perspective. Rational: Many of us who work with X11 on the job would like to have the same work environment at home when we ``dial in''. Prohibiting this are cost and performance considerations. Not many people or institutions can afford to extend the company's network out to people's homes and install workstations there. Alternatively, using high speed leased lines or modems and running high level encapsulated high level network protocols across the link leads to intolerable performance. [For example: currently, even with high speed dedicated lines, attempting to run IP over these relatively slow (compared to LAN speed) lines wastes much of the available bandwidth with IP overhead. Proposed SLIP protocol changes like header compression will help, but reports indicate that interactive use is still no fun. And, it should also be pointed out, high speed leased lines cost ...] Additionally, it would be nice to be able to take advantage of already existing equipment like personal computers for this task. There are also situations at many institutions which duplicate these cost/performance issues on site. For instance, where there's a desire to put a personal computer of some kind and access to a network Window System environment on an employee's desk, but the cost of the network installations, duplicated hardware, etc. is excessive. The goal therefore is to come up with a cost effective solution that let's personal computers be used, where desired, and runs with reasonable performance over low to high speed communications links like EIA232. The situation cries out for a standard ... :-) Language and Terms: Workstation A Display, keyboard and some graphics input device. Network Host Computer system with facilities for interprocess communication. Communications Link For our purposes, any media and link layer protocols connecting a Workstation and Network Host. X11 Server Traditional X11 Server containing Server/Client communications modules, high level graphics facilities, graphics database, and Workstation resource management functions. X11 DIX Server X11 Server whose Workstation resource Xdix management code has been removed. It fulfills it's input and output functions by working with an X11 DDX Server via an XRPCP. X11 DDX Server Workstation management code that runs on a DX``platform'' Workstation. This code provides access to the Ex: DXvga, DXmac Workstation's resources via an XRPCP. For purposes DX* of brevity we'll often refer to X11 DDX server(s) in the abstract as ``DX*''. Design: It's impossible to consider the above outlined design without addressing the alternatives. In particular: Run the X11 Server on the Workstation and come up with a simple, more efficient representation of the normal network protocols for low bandwidth links. For EIA232 links in particular, even if we could come up with a more efficient representation of normal network protocols, it would still have to carry multiplexing and possibly other information that isn't necessary to this application. Certainly this would effectively require that we implement higher level network protocol engines on the Workstations. And which higher level protocols are we going to support? TCP/IP? DECNET? Others? There's also the hassles associated with set up, shut down, and routing of new network links. But most importantly, putting the whole X11 Server on the Workstation would lead to problems with server memory allocation. The server would either end up with a fixed amount of local memory to live in; have to assume local disk to page off of; or support remote paging across the link. Living in a fixed amount of memory is a pain -- when the server runs out of memory, it has to start balancing its resource usage (usually manually unless we want to get into the can of worms presented by resource sharing algorithms). And it should be remembered that many of the Workstations we're interested in supporting won't have much memory. Assuming local disk leaves too many Workstations out in the cold and requires writing paging software. Paging across the link uses up link bandwidth and also requires writing paging software (this time on both ends of the link). Putting most of the X11 Server on the Network Host as an X11 DIX Server takes advantage of the Network Host's native operating system memory management (virtual memory, paging, swapping, etc.). This effectively amounts to paging across the link since redrawing any graphics object requires that information be resent from the Network Host. Some of this load should be alleviated by proper design of an efficient XRPCP and possibly something like X11 save-unders. This also makes effective use of the Network Host's networking facilities and avoids most of the hassles associated with trying to manage dynamic network configurations. On the down side, this will probably lead to higher load on the Network Host both in terms of memory and CPU usage. It's my feeling that putting the most of the X11 Server on the Network Host is the right approach. I believe that the extra load on the Network Host is justifiable since it's likely to have more memory and CPU resources than many of the Workstations that we want to support. Technical Outline: An X11 DDX Server is software that runs on a Workstation. The X11 DDX Server operates in one of two modes: Dumb Terminal or X11 DDX Server. In Dumb Terminal mode, the X11 DDX Server simply passes key strokes through a communications link and displays incoming characters in some simple-minded fashion. In X11 DDX Server mode, the X11 DDX Server provides various graphics facilities via an X11 DIX/DDX RPC Protocol to an X11 DIX Server. Switching from one mode to the other is instigated by the X11 DIX Server. There should be some method to reset the X11 DDX Server to Dumb Terminal mode in the event of errors causing the link to ``lock up''. An X11 DIX Server is software that runs an a Network Host. It provides the standard X11 Server facilities. It implements its X11 Server functionality via the facilities provided by an X11 DDX Server. A typical scenario would have a user at home using a Workstation (e.g. a Macintosh II, an IBM PC with a VGA, an X11 Terminal, or any other hardware capable of running an X11 DDX Server). The user would have a modem attached to the Workstation and use that in Dumb Terminal mode to dial up a remote Network Host. Once logged in to the Network Host, the user would start up an X11 DIX Server, specifying any communication options. The X11 DIX Server would establish ownership of X11 Window System resources (X11 Display number, etc.), and then commence negotiation with the X11 DDX Server to establish an X11 DIX/DDX RPC Protocol connection and cause the X11 DDX Server to enter into X11 DDX Server mode. X11 applications would then connect to the X11 DIX Server in the same way that they would connect to any traditional X11 Server. When the X11 DIX Server shut down, it would cause the X11 DDX Server to re-enter Dumb Terminal mode. Protocol Technical Specification Requirements: The following requirements are laid down in an effort to promote flexibility of future protocol changes and operating environments. Initial protocol start up must include XRPCP version negotiation between X11 DIX and X11 DDX Servers. X11 DIX and X11 DDX Servers will exchange supported protocol versions and agree on a mutually supported version. If no version is supported by both, the X11 DIX Server will exit with an error. It may be necessary to specify a minimal set of versions that must be implemented by all X11 DDX Servers and X11 DIX Servers, but I don't think that this would be necessary or wise. Market presure should cause implementors to make most popular versions available without the need to institutionalize the dreaded disease of backwards compatibility. As an example, we don't require current X11 Servers to support older versions of X protocols ... In the case of multiple mutually supported protocols, a decision must be made as to which to use. This draft does not address that issue beyond mentioning some possibilities. It might be presumed that a higher version number would indicate an improvement over earlier versions. Therefore the highest mutually supported version should be used. Another possibility might be that certain versions are simply tuned for various communications media. If that is the case, there should probably be a mechanism to select a particular version over other mutually supported versions. An example of such a biasing might be starting up an X X11 DIX Server via: Xdix -XRPCP 5 The following diagram gives a description of the communications arrangements over various media using ISO terminology: 7. APPLICATION: XRPCP 6. PRESENTATION: byte ordering, compression, security 1-5. communications: point to point: EIA232, etc.: 5. SESSION: NONE 4. TRANSPORT: NONE 3. NETWORK: NONE 2. LINK: error detection/correction 1. PHYSICAL: RS232, etc. multicast: TCP/IP over ethernet, etc.: 5. SESSION: TCP 4. TRANSPORT: TCP 3. NETWORK: IP 2. LINK: 802.3, etc. 1. PHYSICAL: ethernet, etc. A PRESENTATION layer is noted in the above diagram that hasn't been covered so far. Obviously any presentation layer services should be negotiated between the X11 DIX and X11 DDX Servers (including of course implementation version agreement as above for the XRPCP), but it's not completely clear what should be in this layer. Certainly byte ordering, but what about data compression and security features? There's also the old ISO vs. non-ISO (typically TCP/IP) argument: why divide one protocol into two protocols simply for layering purposes when it introduces yet another protocol start up and shut down? In the ISO world the APPLICATION and PRESENTATION layers above would almost certainly be introduced as two orthogonal network protocols. In the TCP/IP world they would probably be integrated into a single protocol. Since my experience is predominately in the TCP/IP world and I am an adherent to its principles, I prefer to think of the two layers as simply two different functional sections of one inseparable protocol all of whose start up, option and version negotiation, and shut down are bundled in one network layer connection. Practical Considerations: The initial XRPCP and the negotiation mechanisms to start up and shut down an XRPCP connection must be designed. Implementation of error correction and security protocols can be deferred for later versions. I propose simply using most significant byte first for byte ordering in the first version. Later versions might use byte ordering agreement negotiation. A sample X11 DIX Server, Xdix, must be designed and implemented. At least one X11 DDX Server must be designed and implemented. The first X11 DDX Servers should probably be for one of the popular personal computers like the Macintosh II or the IBM AT with VGA, DXmac and DXvga respectively. The design of the protocols and negotiation mechanisms will require input from various experts and concerned parties, coordination of draft proposals, and reviews of those drafts. Protocol design issues will be simplified by the fact that the protocol is essentially going to be a direct implementation of the X11 DIX/DDX interface with a remote procedure call mechanism instead of the more typical procedure call interface. This can be implemented using standard RPC methods of stub procedure interfaces, etc. The protocol design will be complicated be any data sharing that may exist between the DIX and DDX sections of the X11 server. Such data sharing will require additional facilities in the DDX RPC stub section of the X11 DIX server to enable the X11 DDX Server to access the shared data. (Yes, I'm proposing that any such shared data be kept in the X11 DIX Server since this will entail the fewest modifications to the present X11 DIX code.) I'm not familiar with the X11 Server, so I don't know how extensive data sharing such as this may be. Implementation of a sample X11 DIX Server and X11 DDX Servers will require donations of time from various individuals for coding and testing, and coordination of those efforts. One possibility is to seek donations of already existing protocols and code as a starting point. (A candidate that comes immediately to mind for X11 is GraphOn. Something could probably be learned from AT&T's Blit/Layers design also.) This would require convincing such companies that it is in their best interest to do this. Possible arguments are: 1. This standards effort will go on whether they participate or not. It's in their best interest to have input on the process, possibly including large scale adoption of their protocols. 2. There will still be a market for high performance implementations of X11 DIX and X11 DDX Servers for various platforms, just as there is for more traditional servers. There will also still be a market for terminals that implement X11 DDX Servers. One could argue in fact that the market will be larger because of standardization. Summary: I believe that it is both possible and desirable to design a standard for cost effective and reasonable performance support of the X11 Window System over low bandwidth communications links. I believe that such a standard would benefit the user community and open a business market that is only now getting started. If no one else more capable is willing, I volunteer to coordinate the design and implementation of this protocol, and whatever coding and testing I am capable of. Casey