mcdermot@unmvax.UUCP (04/26/85)
A friend recently brought up ultrix-11 on a microPDP11 with
an 11/73 processor. He claims that with 2 users running the
system is really slow. DEC support told him that ultrix-11
has NO BUFFERING on terminal I/O (that is, characters are
all output as they are sent to the driver). And that there
is no buffering in the usual stdio routines.
I don't have the source, but this seems a bit strange to me.
The 2.9 system here does not have these "features" nor did
our old 2.8.
Any ideas/remarks/truths?
john
--
John McDermott {gatech|ucbvax|convex|pur-ee}!unmvax!mcdermot
Univ of NM W (505) 277-4650
Albuquerque, NM 87131 H (505) 255-7796dan@rna.UUCP (Dan Ts'o) (04/29/85)
> A friend recently brought up ultrix-11 on a microPDP11 with > an 11/73 processor. He claims that with 2 users running the > system is really slow. DEC support told him that ultrix-11 > has NO BUFFERING on terminal I/O (that is, characters are > all output as they are sent to the driver). And that there > is no buffering in the usual stdio routines. Yes most older versions of UNIX did not buffer tty output from the stdio package, although the kernel stashes write requests in an output queue. Tty stdio buffering just reduces system call overhead and in some cases, hardware setup overhead. I believe that ULTRIX-11 is based on V7M, which might be a little slower than VENIX-11, PWB/Unix or 2.9BSD. But my best guess at the reason that your friend's 11/73 system is so slow is the Micro11's disk, the RD51/2 with the RDQX controller. This disk subsystem has got to be the worst performer in DEC's history (although it is the cheapest DEC disk ever). I would serious consider an RX02 floppy to be a performance rival. The RDQX controller has a T-11 chip (of Falcon 11/21 fame (a real performance dog)) and it literally takes seconds to respond. Otherwise the 11/73 is quite a speedy CPU. It is just under an 11/44 and should be able to support say ten average users if it had enough memory and a good disk. I put an Eagle on our 11/73 and it hums right along (running a version of PWB/Unix.) Another site I know of uses an 11/73 with a CDC 9448 and VENIX-11, and it is also quite a speedy system -- runs with six users all the time (its all there is) with no response time problems. In a series of UNIX system benchmarks, our 11/73 with the Eagle varied between .14 and .67 of an 11/780 (w/ FPA, Eagles and 4.2BSD), averaging .37 The largest differences (.14, .16) were due to floating point and file copying (thanks to 4.2BSD's fast filesystem) and the smallest different .67 was running the integer sieve benchmark. The 11/73 nroff benchmark ran at .34 of the 11/780. So crudely speaking, if your 11/780 can support 30 users, the 11/73 should be able to support at least ten users. Cheers, Dan Ts'o Dept. Neurobiology Rockefeller Univ. 1230 York Ave. NY, NY 10021 212-570-7671 ...cmcl2!rna!dan
sam@wdl1.UUCP (Sam Kimery) (05/03/85)
> A friend recently brought up ultrix-11 on a microPDP11 with > an 11/73 processor. He claims that with 2 users running the > system is really slow. DEC support told him that ultrix-11 > has NO BUFFERING on terminal I/O (that is, characters are > all output as they are sent to the driver). And that there > is no buffering in the usual stdio routines. > Is the system running a combination of DZ and DH muxes? DEC tech. support claims that the two together will bring the 73 and the micro 11 to its knees. I know for a fact that the 73 is a stiff processor. I used to run on a 73 with 20 (my own mod. unsupported by DEC) users, and had better response than the 70 that I am running on now. Sam
martillo@mit-athena.UUCP (Joaquim Martillo) (05/07/85)
Console I/O is much less efficient the ordinary terminal I/O. If your friend only has a console on his pdp11, he will find I/O very slow.
jih@usl.UUCP (Juha I. Heinanen) (05/09/85)
In article <821@unmvax.UUCP> mcdermot@unmvax.UUCP writes: >A friend recently brought up ultrix-11 on a microPDP11 with >an 11/73 processor. He claims that with 2 users running the >system is really slow. Two/three years ago we used to run the predecessor of Ultrix-11, then called Unix V7m, on out PDP-11/34 and we found it was *much* slower than the European Unix Users Group port of the same thing. We never figured out exactly why this was the case, but it must have something to with the way the kernel overlays are done. Anyhow, I would like to hear from somebody at the DEC Unix support group why DEC has not based their Ultrix-11 on 2.9BSD which is even much faster that the EUUG port. -- Juha Heinanen USL, P.O. Box 44330, Lafayette, LA 70504-433, tel. (318)231-5345 UUCP: {ut-sally, akgua}!usl!jih ARPA: usl!jih@ut-sally
dan@rna.UUCP (Dan Ts'o) (05/09/85)
In article <> martillo@mit-athena.UUCP (Joaquim Martillo) writes: >Console I/O is much less efficient the ordinary terminal I/O. >If your friend only has a console on his pdp11, he will find I/O >very slow. It is not clear what you mean... If you mean that output generated from the kernel version of printf() in older UNIX's usually in the form of error messages from the kernel printed on the console is very slow and efficient, then you're right. The internal version of printf() does not use interrupts and character buffers, but tight loops on console output which pretty much suspends the rest of UNIX. This "feature" has been fixed in some UNIX's. However, it is not the normal mode of "console I/O" and could not be considered a reason why ULTRIX-11 is slow. If you mean that console I/O is slow because on PDP-11's it is usually done to DL-11 serial interfaces which have no silo nor DMA capability, in contrast to terminals on DH-11's and DMF-11's. This situation is also true but I think negligible in most cases. We run 11/23's and 11/73's on DL-11 interfaces and aren't bothered too much by the performance of the DL-11's. But if you're doing heavy I/O on the console line, e.g. graphics or UUCP, then it might be a problem. If you mean that older UNIX's don't buffer stdout on terminals, that's right but valid for all serial lines. This problem is much more of a performance factor than the DL-11 problem.
jack@boring.UUCP (05/13/85)
In article <490@usl.UUCP> jih@usl.UUCP (Juha I. Heinanen) writes: >Two/three years ago we used to run the predecessor of Ultrix-11, then >called Unix V7m, on out PDP-11/34 and we found it was *much* slower than >the European Unix Users Group port of the same thing. We never figured >out exactly why this was the case, but it must have something to with >the way the kernel overlays are done. I always read this remark about v7m, but I think it is just not true. If you massage the overlaying by hand you can get a very reasonable performance out of it. We're running an 11/34 with 10 users, quite comfortably (I should tell that we have an EMMU, 1Mb of mem, and a Capricorn on it, but still). Also, if you're not satisfied with performance, you could hack some things like XBUF mod, FIFO inode allocation, etc. in from the EUUG distribution, and still have an overlay kernel. -- Jack Jansen, jack@mcvax.UUCP The shell is my oyster.
borman@decvax.UUCP (Dave Borman) (05/28/85)
Some information on ULTRIX-11, DEC's version of UNIX* for the PDP-11. The most common question asked is "Why isn't ULTRIX-11 based on 2.9BSD?" Version 1.0 of ULTRIX-11 was called V7M-11, and was released in August of 1983. When the work on it was being done, 2.9BSD was not available. By the time we got a real 2.9BSD tape we were well into development of 2.0, and it would have been foolish for us to put all our changes into 2.9BSD (The time to put in the changes and verify everything would have meant that 2.0 would still not be out the door). ULTRIX-11 has followed a similar history to 2.9BSD, just not at the same pace. It's not based on 2.9BSD, but it has many 2.9BSD features. For example: job control, the terminal driver, the C-shell, overlayed kernels, and overlayed user processes, to name a few. We have things that 2.9BSD does not, like error logging (keeps DEC field service happy!) floating point simulator in the kernel, system verification programs, new device support, working backup code for all supported processors. I can just about guarantee that ULTRIX-11 has the best MSCP (ra) driver you'll find for the PDP-11. Sadly enough, some things did not get into 2.0, like stdio line buffering. This is probably the one missing feature that gives the greatest appearance of ULTRIX-11 being slow. This is regretable, but schedules did not allow enough time. It will be in the next release, along with many more features and performance enhancements: 1K file system, System V compatability, 4.2 sockets/TCP/IP (4.2 sockets you say? and running on an 11/23+ with .5 meg memory? I did the work, I'll be at Portland USENIX, and will be glad to answer questions. We should have a pre-release running in the booth). -Dave Borman, Digital UNIX Engineering Group decvax!borman * UNIX is a trademark of AT&T Bell Labs * ULTRIX is a trademark of Digital Equipment Corporation