Re: [Simh] Reference performance benchmarks
On 2016-12-12 23:50, Jacob Goense wrote: On 2016-12-12 16:34, Johnny Billquist wrote: I would start at CPU idle detection. It seems to be working, but definitely a good place to look at to get more bang(vaxen) for my bucks(cheap x86's/rpi clones). You said simh, running in a browser. How much CPU is the browser using if the machine in simh is doing nothing? Not to mention if you happen to run two instances? Johnny -- Johnny Billquist || "I'm on a bus || on a psychedelic trip email: b...@softjar.se || Reading murder books pdp is alive! || tryin' to stay hip" - B. Idol ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
On 2016-12-12 16:34, Johnny Billquist wrote: I would start at CPU idle detection. It seems to be working, but definitely a good place to look at to get more bang(vaxen) for my bucks(cheap x86's/rpi clones). ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
On 2016-12-12 16:09, Jeremy C. Reed wrote: For VAX-11/780 see the benchmark results and source code from Kashtan's research. AUUGN-V02.4.pdf starting on PDF page 24. And see Joy's followup in AUUGN-V02.3.pdf (PDF Page 60). See http://minnie.tuhs.org/Archive/Documentation/AUUGN While your OS may be different it has numbers for some different systems so may give you some rough estimates. If this code isn't online, it would be nice to make available (maybe put on github and let us know). Oh, I see how I can turn that into a canary and let Travis have a ball with it. This is a great help. (Hopefully someday soonI will get my book completed that has this story.) I think I got a sneak preview of the 386BSD patchkits chapter through the unintended wonders of google cache once. I'm a sure customer and will probably empdapt[1] some of it. [1] emdaptation [em-d-uh p-tey-shuh n] noun 1. the state of being emdapted; running in an emulator. 2. something produced by emulating: an emdaptation of a UNIX history book. Origins: Word play on emscripten, embiggen, emulating, adapting ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
On 2016-12-12 16:29, khandy21yo wrote: Is your speed problem limited to disk access, or is it also CPU related? Running in a browser might mean that the disk access is going through the network, which can slow it down excessively, There are several bottlenecks on a system, like slow disks, that don't effect other aspects, like instruction speed, it's important to determine exactly where the speed limits exists. CPU instruction speed, excessive swapping, disk bandwidth, terminal baud water, . .. They can all be a problem, and it can be more than one at a time. For the browser case I have the feeling it is guest device I/O from one to the other that is taking way too long, but the real problem is that I'm not able to gauge if I am going way over time accurate speeds and even want to bother optimising. For the case where I am spawning vaxen on a small small box I'm CPU bound host wise, and not sure when to stop. ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
I would start at CPU idle detection. Johnny On 2016-12-12 22:29, khandy21yo wrote: Is your speed problem limited to disk access, or is it also CPU related? Running in a browser might mean that the disk access is going through the network, which can slow it down excessively, There are several bottlenecks on a system, like slow disks, that don't effect other aspects, like instruction speed, it's important to determine exactly where the speed limits exists. CPU instruction speed, excessive swapping, disk bandwidth, terminal baud water, . .. They can all be a problem, and it can be more than one at a time. Sent from my Galaxy TabĀ® A Original message From: Jacob Goense Date: 12/12/16 12:13 PM (GMT-07:00) To: simh@trailing-edge.com Subject: [Simh] Reference performance benchmarks I'm running simh in rather tight corners. Mostly VAX/780 and PDP-11/70 with V6 or BSD. Simh runs in a browser, or, I have (too) many of them on underpowered Asus Eee PC's. Any tips on how I can check if I have taken things too far? Like minimum values to look for in SHOW CLOCK, or any tools I can run that can be referenced against same on real kit. I run into the same question on the I/O end. During the installation of v6 I can do dd if=/dev/mt0 of=/dev/rk1 bs=2048 count=1000 skip=1025 and it is done in a second on machine that is doing nothing else. With simh running in a browser on a small busy machine it can take up to an hour. I have no idea how to gauge how far this is off from real kit. ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh -- Johnny Billquist || "I'm on a bus || on a psychedelic trip email: b...@softjar.se || Reading murder books pdp is alive! || tryin' to stay hip" - B. Idol ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
Is your speed problem limited to disk access, or is it also CPU related? Running in a browser might mean that the disk access is going through the network, which can slow it down excessively, There are several bottlenecks on a system, like slow disks, that don't effect other aspects, like instruction speed, it's important to determine exactly where the speed limits exists.CPU instruction speed, excessive swapping, disk bandwidth, terminal baud water, . .. They can all be a problem, and it can be more than one at a time. Sent from my Galaxy TabĀ® A Original message From: Jacob Goense Date: 12/12/16 12:13 PM (GMT-07:00) To: simh@trailing-edge.com Subject: [Simh] Reference performance benchmarks I'm running simh in rather tight corners. Mostly VAX/780 and PDP-11/70 with V6 or BSD. Simh runs in a browser, or, I have (too) many of them on underpowered Asus Eee PC's. Any tips on how I can check if I have taken things too far? Like minimum values to look for in SHOW CLOCK, or any tools I can run that can be referenced against same on real kit. I run into the same question on the I/O end. During the installation of v6 I can do dd if=/dev/mt0 of=/dev/rk1 bs=2048 count=1000 skip=1025 and it is done in a second on machine that is doing nothing else. With simh running in a browser on a small busy machine it can take up to an hour. I have no idea how to gauge how far this is off from real kit. ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
On Mon, 12 Dec 2016, Jacob Goense wrote: > >> On Dec 12, 2016, at 2:13 PM, Jacob Goense wrote: > >> I'm running simh in rather tight corners. Mostly VAX/780 and PDP-11/70 > >> with V6 or BSD. Simh runs in a browser, or, I have (too) many of them > >> on underpowered Asus Eee PC's. Any tips on how I can check if I have > >> taken things too far ... > want to be able to say that it is not that far off from the original > speed. I'm relatively young (from 1975) and lack the experience to > have even the faintest clue what these speeds are. For VAX-11/780 see the benchmark results and source code from Kashtan's research. AUUGN-V02.4.pdf starting on PDF page 24. And see Joy's followup in AUUGN-V02.3.pdf (PDF Page 60). See http://minnie.tuhs.org/Archive/Documentation/AUUGN While your OS may be different it has numbers for some different systems so may give you some rough estimates. If this code isn't online, it would be nice to make available (maybe put on github and let us know). (Hopefully someday soonI will get my book completed that has this story.) ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
> On Dec 12, 2016, at 3:23 PM, Jacob Goense wrote: > > ... > For development purposes I'd run a pdp11 simh on a vax simh in a > browser, wait hours for it to get something done and find that more > than reasonable. Just that, when opening these Droste effects of > emulators in emulators in emulators in browsers to a wider public, I > want to be able to say that it is not that far off from the original > speed. I'm relatively young (from 1975) and lack the experience to > have even the faintest clue what these speeds are. "Droste effect", indeed. Instruction timings of various PDP-11 models are documented in detail in the various Processor Handbooks. For example, on the PDP-11/20 a MOV takes 3.2 microseconds, an ADD 4.8 -- plus additional time for some addressing modes. I don't remember whether the same applies to VAX. Very roughly an 11/780 was generally viewed as a one MIPS machine. (The same goes, again roughly, for fast PDP-11s like the 11/70.) In general SIMH won't accurately match the timing details of the real machine, because individual instructions will take time according to the complexity of the emulation rather than that of the implementation. An accurate model of a non-IEEE floating point instruction is probably comparatively slow. But a shift might be fast (if the original machine shifted serially while the new one does it with a barrel shifter). paul ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
On 2016-12-12 14:34, Paul Koning wrote: On Dec 12, 2016, at 2:13 PM, Jacob Goense wrote: I'm running simh in rather tight corners. Mostly VAX/780 and PDP-11/70 with V6 or BSD. Simh runs in a browser, or, I have (too) many of them on underpowered Asus Eee PC's. Any tips on how I can check if I have taken things too far? Depends on what you mean by "too far". Do you want SIMH to run as fast as, or faster than, the real hardware? If so, running, say, the PDP11 simulator on a 780 is probably "too far". But it may be ok for the 1620 emulator. If by "too far" you mean "not useable", that depends on your tolerance. I don't mind if things run faster. In case a user runs into usability issues I want to be able to gauge whether it is because simh is running significantly slower than the real hardware or just time accurate. People have run PDP-11 emulators on a PDP-10 (MIMIC). There's a document describing an Electrologica X8 emulator that runs on its predecessor the X1 -- a machine with no floating point hardware and a typical instruction time around 50 microseconds. So the emulation runs at about 4-8 ms per emulated integer instruction, substantially slower for float. But it was (apparently) used for initial development of substantial programs such as an ALGOL compiler. It's all a question of what you consider reasonable. For development purposes I'd run a pdp11 simh on a vax simh in a browser, wait hours for it to get something done and find that more than reasonable. Just that, when opening these Droste effects of emulators in emulators in emulators in browsers to a wider public, I want to be able to say that it is not that far off from the original speed. I'm relatively young (from 1975) and lack the experience to have even the faintest clue what these speeds are. ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh
Re: [Simh] Reference performance benchmarks
> On Dec 12, 2016, at 2:13 PM, Jacob Goense wrote: > > I'm running simh in rather tight corners. Mostly VAX/780 and PDP-11/70 > with V6 or BSD. Simh runs in a browser, or, I have (too) many of them > on underpowered Asus Eee PC's. Any tips on how I can check if I have > taken things too far? Depends on what you mean by "too far". Do you want SIMH to run as fast as, or faster than, the real hardware? If so, running, say, the PDP11 simulator on a 780 is probably "too far". But it may be ok for the 1620 emulator. If by "too far" you mean "not useable", that depends on your tolerance. People have run PDP-11 emulators on a PDP-10 (MIMIC). There's a document describing an Electrologica X8 emulator that runs on its predecessor the X1 -- a machine with no floating point hardware and a typical instruction time around 50 microseconds. So the emulation runs at about 4-8 ms per emulated integer instruction, substantially slower for float. But it was (apparently) used for initial development of substantial programs such as an ALGOL compiler. It's all a question of what you consider reasonable. paul ___ Simh mailing list Simh@trailing-edge.com http://mailman.trailing-edge.com/mailman/listinfo/simh