gdburns@TBAG.OSC.EDU (Greg Burns) (02/03/90)
As promised, here is information about Trollius(TM) 2.0 and the project in general. My apologies for the duplicate message to members of both lists. <<<To get to the bare facts, search for FACTS>>> Though we've been fairly quiet on the transputer mailing list, a list that we started shortly after Trollius was started, our transputer activity has not skipped a beat. It may skip several beats in the near future, as we get ready for an i860 port. We still think transputers are cool and we are hoping, like most of you, for faster devices. We'll try to keep the list better informed, for Trollius is not proprietary software. In 1989, we completed an overhaul of Trollius 1.1, adding several features, streamlining others, and building a new layer on top, the topology layer. The guiding principle of the Trollius overhaul was portability. If you recall Trollius postings of 1987, the first version of the software was targetted at the FPS T-Series (hereafter, never to be mentioned again). We went to a LOT of trouble to support that particular set of schematics. Trollius 2.0 is targetted at a very general multicomputer model. We don't specifically support any particular hardware, though the software was developed on a NiCHE NT1000. The NT1000 is a nice 9u VME motherboard, but at the end of the day, it is just another bag of transputers (hence tbag). What sets it apart is its clean and simple device driver and software configuration. You're not going to find a better generic UNIX to transputer software interface - just what we system programmers love. Trollius 1.1 was portable, provided you had a guru to do the port. We did port 1.1 to the Topologix T1000 (and they renamed it LogixOS), and the NiCHE NT1000 (and they renamed it PRE, now Transtech is planning to call it Genesys -- you can't tell the Trollius players without a programme). Given a device driver that doesn't go out of its way to make life difficult, anybody can port 2.0. The Trollius multicomputer model emphasizes the basics. You have processors with local memory running processes and communication links, possibly software reconfigurable and everybody gets along with messages. A node is a transputer, a Sun, a Cray Y-MP, u-name-it. You choose the nodes and the interconnections and Trollius configures, boots and routes. The philosophy is similar to that espoused by the recently developed Surfaceware (CS Tools) by Meiko. Hmmm. But while many systems only persuade you to use 2,4,8 simple calls and you needn't worry about nodes, processes and messages, Trollius also wants these objects to kneel before the creative developer and say, "Oh Captain, my Captain". We do have a system with a big button that says PLAY (2 calls actually) but we also have lots of smaller buttons for the researchers and the experimentalists. Here is a list of the new or improved buttons in Trollius 2.0. FACTS ** COMPLETE TOPOLOGY INDEPENDENCE: User creates a boot schema (our fancy name for a config file) that specifies nodes, nodeids, interconnections, and optionally, communication loads, explicit routes, route priorities. Trollius uses the boot schema to configure the links (if necessary), create routing tables and boot the system. System administrators can install default boot schemata as necessary so users don't have to write them. ** UNIX STYLE SIGNALS: This is not a hook into UNIX. Trollius has its own signals using the UNIX standard interface. You can stop a process and then continue it. There are a couple of user defined sigs that you can catch. ** MULTI-PRIORITY SCHEDULING: Above and beyond the underlying scheduler, you can optionally specify a Trollius scheduling priority on a "highest pri process runs till blocked" strategy. Allows you to reliably predict the relative behaviour of processes when only two transputer priorities cramps your style. ** MULTIPLE VIRTUAL CIRCUITS: For low overhead IPC, a virtual circuit bypasses the kernel and establishes an underlying comm. channel with another process. In the case of a transputer link process, it means you drive the link directly. You can now have up to 16 of these simultaneous circuits, start with a dynamic system, end up with a static Occam-like system and never leave the standard Trollius calls. ** RESIDENT SHARED LIBRARY: This means that 15 Trollius processes don't have to duplicate the IPC library. We're going for 64K of code/data/stack. ** DECREASED IPC SETUP TIME: improved at all levels for both host and compute nodes. We will post numbers. ** MULTICASTING/MULTIREELING: Fan it out and combine it back for any source node and any set of leaf nodes. Use builtin combining functions for reeling or supply your own. (experienced Trollius users: tping N, cast/reel echo to all nodes, very cool) ** NEW BUFFERING: Look at those messages that aren't being received. You get to tune space allocation and throughput capability - much less prone to link jamming - a 1.1 problem. ** EXPANDED DOCUMENTATION: more information on how to use and program Trollius for both C and Fortan ** PARALLEL FILE ACCESS MODES: We've added O_SINGL and O_MULTI flags to the open call that invoke CUBIX (invented at Cal Tech) functionality. ** CrOS III: (invented at Cal Tech) library from Fox's book ** BRENDA: This was available for 1.1 and has been polished for 2.0. Brenda provided the distilled functionality of Linda (invented at Yale) and is very handy for farmer/worker problems. ** LOGICAL SYSTEMS TRANSPUTER TOOLSET: It produces good code, has a blazing math library, is a cross development kit, has inexpensive source code and has a development future, including Fortran. ** STAND-ALONE DEVELOPMENT: We emulate multicomputer node identifiers on a single UNIX host node and the natural Trollius multitasking node environment does the rest. We highly recommend this as a first step in the debugging process. The above list covers new features only. We'll review other Trollius features on the trollius mailing list. There will be a third chapter to Trollius. Trollius is a trademark of The Ohio State University Research Foundation and the Cornell University Research Foundation. -=- Greg Burns gdburns@tbag.osc.edu Trollius Project (614) 292-8492 Research Computing The Ohio State University Trollius - hand made - fine tailoring
jeremy@cs.ua.oz.au (Jeremy Webber) (02/06/90)
So, Where D'ya Get It? -jeremy -- -- Jeremy Webber ACSnet: jeremy@chook.ua.oz Digital Arts Film and Television, Internet: jeremy@chook.ua.oz.au 60 Hutt St, Adelaide 5001, Voicenet: +61 8 223 2430 Australia Papernet: +61 8 272 2774 (FAX)