dem@uwslh.UUCP (David E. Miran) (01/31/86)
I recently received a brochure from Avalon Computer Systems for
their AP/10A attached processor system.
It is an auxiliary cpu that plugs into the VAX unibus.
You add a device driver and recompile your programs so that they
will be loaded into the new cpu rather than run by the VAX cpu.
It apparently runs only one process at a time in the attached processor.
It is supposed to run somewhat faster than a VAX-11/750.
Supposedly you can attach several of them to a VAX.
At about $7,000 for one and quantity discounts it sounds promising.
Has anyone actually tried one of these? Is it worth the trouble?
How do you deal with someone trying program1|program2|program3
where more than one of the programs has been compiled to run on the
attached processor.
--
David E. Miran ...!{seismo,harvard,topaz,ihnp4}!uwvax!uwslh!dem
Wisconsin State Hygiene Lab (608) 262-0019
University of Wisconsin
465 Henry Mall
Madison, WI 53706wcs@ho95e.UUCP (x0705) (02/01/86)
In article <135@uwslh.UUCP> dem@uwslh.UUCP writes: >I recently received a brochure from Avalon Computer Systems for >their AP/10A attached processor system. >It is an auxiliary cpu that plugs into the VAX unibus. >You add a device driver and recompile your programs so that they >will be loaded into the new cpu rather than run by the VAX cpu. Last time I looked at them, they only supported 4.*BSD, and not System V. Does anybody know of a System V version? >It apparently runs only one process at a time in the attached processor. >How do you deal with someone trying program1|program2|program3 >where more than one of the programs has been compiled to run on the >attached processor. If your program is written properly, it will check a file or other semaphore mechanism, then run either the coprocessor version or the main CPU version depending on what's free. >Has anyone actually tried one of these? Is it worth the trouble? We have a similar device that lives out on an RS232 port and does troff. It's hard to say if it's worth while, aminly because of the time to transmit the files down to the box and back. If it were on a network that didn't require too much CPU to transfer files, it would be a big win. -- # Bill Stewart, AT&T Bell Labs 2G-202, Holmdel NJ 1-201-949-0705 ihnp4!ho95c!wcs
bhl@uclachem.UUCP (Brad Lowman) (02/08/86)
In article <135@uwslh.UUCP> dem@uwslh.UUCP writes: > I recently received a brochure from Avalon Computer Systems for > their AP/10A attached processor system. > [...] > At about $7,000 for one and quantity discounts it sounds promising. > Has anyone actually tried one of these? Is it worth the trouble? A number of us recently went to Avalon's Glendale office to speak with a principal of their firm about using the AP/10A system. Their brochure will tell you of all the advantages to this system. I will list a few concerns/ disadvantages here: Since all I/O is priviledged on a VAX, the AP/10A must sent a packet, via the Unibus, to the OS requesting the I/O operation to be performed. Once completed, another packet must be sent back to the AP/10A with the results of the operation. This means the cost of doing IO is now 3 system calls instead of 1. You must have the complete source for any program you wish to run on the AP/10. Avalon distributes the standard Unix libraries compiled for use on its product, but the CPU-bound jobs that work so well on the AP/10A tend to our third-party software packages, for which we don't (or can't) get the sources for. Since only one process can run on the AP/10A at a time (all others are blocked), this process can not fork, although I believe exec'ing another AP/10A-compiled process is allowed. In summary, if you have a fairly simple CPU-bound f77 or C applications that take less than 11.5 Meg virtual memory, this board is worth looking into. You don't want to put highly IO-bound, especially terminal IO-bound, programs like "vi" or some database programs on it though. --brad lowman UUCP: {ucbvax!ucla-cs,ihnp4!bradley!cepu}!uclachem!bhl AT&T: (213) 825-1824 ARPA: uclachem!bhl@locus.ucla.edu