root@ucbvax.UUCP (09/09/83)
From GEOFF5@SRI-CSL Fri Sep 9 10:14:12 1983 I asked several days ago if anyone had ideas why my Versatec printer/ plotter (heavy character IO) runs faster on a 730 than on the 780. I got 3 responses all suggesting that it is the DMF32 which runs faster than a DZ11. I appreciate these responses, however the Versatec does not run off a serial port, it has its own parallel interface (which can run DMA or character - senario occurs in character mode). Thus the DMF32/DZ11 difference does not explain it. One response also mentioned the 730 Unibus goes 20% faster than the 780's. Is this true? I thought the Unibus was clocked at some fixed speed? Another idea I had is that the Unibus is connected more difectly to the 730 and so maybe there is one less layer of software to go through. (I think interrupts are directly vectored as in the 750 for example.) Anyone know what is really going on? Rg -------
daemon@ucbvax.UUCP (09/15/83)
From GEOFF5@SRI-CSL Thu Sep 15 00:09:41 1983 Date: Wed 31 Aug 83 23:57:33-EDT From: Richard Garland <G.GARLAND@COLUMBIA-20.ARPA> Subject: 730 Unibus speed revisited. I asked several days ago if anyone had ideas why my Versatec printer/ plotter (heavy character IO) runs faster on a 730 than on the 780. One response ... mentioned the 730 Unibus goes 20% faster than the 780's. Is this true? I thought the Unibus was clocked at some fixed speed? One of the nice things about the UNIBUS from the interface designer's perspective, is that it is asynchronous, i.e. each transaction consists of discrete steps that go through a handshake sequence before they proceed. In general, synchronous buses such as the VAX/780 SBI are considered faster than asynchronous buses such as the UNIBUS or the IBM 360/370/4300 data channel. There are however some design guidelines about what expectations you can make about UNIBUS timing. These used to be published in something called the UNIBUS HANDBOOK and the PDP-11 INTERFACING HANDBOOK, but I hear that such information has not been easily accessible in recent years, and recent DEC CPUs have not adhered to the OLD specs. Recent shortenings of timing windows took place with 11/44 and 11/750, so it is not surprising if the 11/730 is shaving off yet another little bit, since it is so dependent on its UNIBUS. However, the important part is not the UNIBUS speed, but the number of instructions that must be executed for each character interrupt (see below). Another idea I had is that the Unibus is connected more directly to the 730 and so maybe there is one less layer of software to go through. (I think interrupts are directly vectored as in the 750 for example.) The VAX HARDWARE HANDBOOK 1982-83 describes the UNIBUS adapter of the 11/730 in just about the same terms as the 11/780. But you really ought to write a device driver that will drive your VERSATEC in DMA mode.... That should make your paper roll flow rather than crawl!! med venlig hilsen Lars Poulsen <lars@ACC>
aps@decvax.UUCP (Armando P. Stettner) (09/18/83)
The current VAX Hardware handbooks are not detailed enough to detail the UNIBUS. As I recall, they only discuss the protocol in broad terms. (See the DEC Bus Handbook for more details.) The technical descriptions that are available for the VAX processors provide much more details about the particular implementations. By the way, when you compare the 780 UNIBUS and the 730 UNIBUS and say that they are the same because the handbook describes them in the same terms, remember the above paragraph. Further, the 780 does break up the UNIBUS transactions (SBI interfacing, etc). The 730 UNIBUS implemention is more similar to the PDP-11 implementations with respect to its relationship to the CPU. aps.