| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
| World Community Grid Forums
|
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 14
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Rom Walton has recently posted some of the upgrades under development for BOINC here: http://www.romwnet.org/dasblogce/
People who are always wondering about program optimization will be particularly happy about: BOINC gains CPU capability detection Starting in the next version of the BOINC client we'll be able to detect CPU capabilities. It is important to note that the capability detection is actually done by the operating system and BOINC just queries the operating system for the supported instruction sets. I bring this up because not all operating systems fully support all additional instruction sets supported by the processor. We are being conservative here to avoid illegal instruction exceptions or privileged instruction exceptions. For Windows the following instruction sets or capabilities can be detected: * fpu * tsc * pae * nx * sse * sse2 * sse3 * 3dnow * mmx On Linux we read the data out of /proc/cpuinfo. I still need to write the code for the Mac OS. The processor information will be passed to both the science applications and the scheduling server. The problem with optimizing for a particular processor type is that it can alter the results a small amount, which is a no-no for quorum validation, unless all the results are run on the same processor type. So this gives projects some options. Of course, whether or not a particular optimization is reasonable or not is a different question. But at least the possibility is there. And look at the simplified GUI. Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
That GUI looks sharp!
----------------------------------------Nice to hear of support for optimised science apps. Perhaps one day attached devices will be associated with a workgroup whose member devices all use the same processor (or processors that yield the same result for any given WU). Ad hoc 3 member workgroups could be formed on the fly from the larger workgroup and dissolved when the WU validates. [Edit 1 times, last edit by Former Member at Jun 29, 2006 2:36:50 AM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Not far fetched, but what about lone me with that hyperthreaded pc?
----------------------------------------The consequence of your proposal would be that a WU send to quorum A. would produce a slightly different result from quorum B. How acceptable would that be to the science?
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Awww, lonely Sekerob, don't feel left out. Your HT machine could still crunch using a non-optimised science app. I imagine there would be other hosts who would not be able to run the optimised apps for various reasons, though I must admit I don't know exactly what those reasons might be.
----------------------------------------The science might be able to work with such small variations. As I understand it a result from a Linux host never matches a result from a Windows host but the difference does not impact the science. I bet results from Macs match neither Windows nor Linux results. But the results from 3 Macs match so it's possible to achieve a quorum and declare the result valid. [Edit 1 times, last edit by Former Member at Jun 29, 2006 3:11:11 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Ugh, I'm not impressed. In fact, I'm horrified. Far from being "simple", it has all sorts of mysterious buttons and widgets. I can imagine the troubleshooting: "Click on the funny squiggle. No, not where is says "My projects. The other squiggle. What, you use a different skin? Why, then just click at random and see what happens."
Meh. Let's disassemble it in full. There's a banner, presumably for the project or team logo banner. Fair enough. Although I get the impression that this is for branding, and doesn't change depending on the selected project. The projects are tabbed. Makes sense, can't fault that. Actually, I think these are results - they may be work units. Not a problem, really. But why the cryptic codes in the titles? The progress is almost certainly CPU time, but is represented as wallclock time. That can't be good. Graphics - well, if you must. Now we hit the buttons. Attach Project is simple enough, but has caught out lots of people before now. Then a weird button that might be fullscreen. Ah, the joys of mystery meat navigation! Why tell people when you can make them guess? Moving down, there's more mysterious buttons. A back and play button? Most unlikely. Icons for My Projects, with weird down arrows? Who knows! M? What the frelling smeg is M? Pause is a well known graphic, and Preferences and Advanced View are labelled, thank $DEITY. In summary, they copied a few elements from UD, and did zero usability testing or analysis, and got it disastrously wrong. Lucky it's open source, really. Maybe I can influence the design a little before it completely blows up in our faces. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Ugh, I'm not impressed. In fact, I'm horrified. Far from being "simple", it has all sorts of mysterious buttons and widgets. I can imagine the troubleshooting: "Click on the funny squiggle. No, not where is says "My projects. The other squiggle. What, you use a different skin? Why, then just click at random and see what happens." Didactylos, why spoil good fun with common sense? After they |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Dagorath, as most of the time, Lucky Luck, my alter ego, thought of running CPU-Z to check on the instruction sets in the HT chip.... comes out 3 of the listed one's are in mine.....no feeling of i(n)s(u/o)lation anymore...no P pills required.
----------------------------------------* fpu * tsc * pae * nx * sse * sse2 * sse3 * 3dnow * mmx PS, what was the name of his hoss again?
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The problem with optimizing for a particular processor type is that it can alter the results a small amount, which is a no-no for quorum validation, unless all the results are run on the same processor type. So this gives projects some options. So, It would be simply a matter of matching work units to 'Like' processors, right? Running a WU in its native instruction set WOULD in fact lead to performance gains no? So, when BOINC goes out to fetch a work unit, why not configure the software to say "Hey WCG, here is your last WU. I am an AMD Athlon 64 X2 3800+. Pleaase send me a new WU." Then WCG Could say OK, we have bbfl running amd64x2, we have lawrencehardin running amd64x2, and we have WCGUser99 running amd64x2. Assign the Same WU to these three boxes, and not only will you not have a variation in your results due to CPU instruction set, but the WUs will finish at approximately the same time, unless one of them has a 4800+ Matching WU with similar CPU Speed/Architecture would seem to stabilize the output of WCG right? It would also make my results get validated faster... |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Ugh, I'm not impressed. In fact, I'm horrified. Far from being "simple", it has all sorts of mysterious buttons and widgets. I can imagine the troubleshooting: "Click on the funny squiggle. No, not where is says "My projects. The other squiggle. What, you use a different skin? Why, then just click at random and see what happens." Meh. Let's disassemble it in full. There's a banner, presumably for the project or team logo banner. Fair enough. Although I get the impression that this is for branding, and doesn't change depending on the selected project. The projects are tabbed. Makes sense, can't fault that. Actually, I think these are results - they may be work units. Not a problem, really. But why the cryptic codes in the titles? The progress is almost certainly CPU time, but is represented as wallclock time. That can't be good. Graphics - well, if you must. Now we hit the buttons. Attach Project is simple enough, but has caught out lots of people before now. Then a weird button that might be fullscreen. Ah, the joys of mystery meat navigation! Why tell people when you can make them guess? Moving down, there's more mysterious buttons. A back and play button? Most unlikely. Icons for My Projects, with weird down arrows? Who knows! M? What the frelling smeg is M? Pause is a well known graphic, and Preferences and Advanced View are labelled, thank $DEITY. In summary, they copied a few elements from UD, and did zero usability testing or analysis, and got it disastrously wrong. Lucky it's open source, really. Maybe I can influence the design a little before it completely blows up in our faces. This is some really great feedback. I know Blaine and Kevin are actively looking for feedback, although most of it has been on boinc_dev so far. I'll be sure to repost this on boinc_dev so everybody can see it. I should point out that nothing in the new UI is written in stone. We are still a ways out from completing it. The more feedback we have the better the end product will be. ----- Rom BOINC Development Team |
||
|
|
Johnny Cool
Ace Cruncher USA Joined: Jul 28, 2005 Post Count: 8621 Status: Offline Project Badges:
|
Ugh, I'm not impressed. In fact, I'm horrified. Far from being "simple", it has all sorts of mysterious buttons and widgets. I can imagine the troubleshooting: "Click on the funny squiggle. No, not where is says "My projects. The other squiggle. What, you use a different skin? Why, then just click at random and see what happens." Meh. Let's disassemble it in full. There's a banner, presumably for the project or team logo banner. Fair enough. Although I get the impression that this is for branding, and doesn't change depending on the selected project. The projects are tabbed. Makes sense, can't fault that. Actually, I think these are results - they may be work units. Not a problem, really. But why the cryptic codes in the titles? The progress is almost certainly CPU time, but is represented as wallclock time. That can't be good. Graphics - well, if you must. Now we hit the buttons. Attach Project is simple enough, but has caught out lots of people before now. Then a weird button that might be fullscreen. Ah, the joys of mystery meat navigation! Why tell people when you can make them guess? Moving down, there's more mysterious buttons. A back and play button? Most unlikely. Icons for My Projects, with weird down arrows? Who knows! M? What the frelling smeg is M? Pause is a well known graphic, and Preferences and Advanced View are labelled, thank $DEITY. In summary, they copied a few elements from UD, and did zero usability testing or analysis, and got it disastrously wrong. Lucky it's open source, really. Maybe I can influence the design a little before it completely blows up in our faces. This is some really great feedback. I know Blaine and Kevin are actively looking for feedback, although most of it has been on boinc_dev so far. I'll be sure to repost this on boinc_dev so everybody can see it. I should point out that nothing in the new UI is written in stone. We are still a ways out from completing it. The more feedback we have the better the end product will be. ----- Rom BOINC Development Team Yes, Didactylos has posted the reality here, and I fully agree with him. Spoken (or posted) like a true UI/Human Factors "technoid". Bloatware is simply that. I know a well-known CTO who lives here in Colorado Springs who used to work for Microsoft. He was one of the original Team Members who had worked for the original MS NT Team. He has authored many books about 3-D graphics, was a co-inventor of AVI for Windows, and has a website about his experiences working there: What is Engineering? Anyone that has heard me give a talk or just sat around a table with me over lunch will have heard me go on and on about how software programming is not yet an engineering subject and software developers are not engineers. This is how I define engineering: I take two napkins from the lunch table. One the first napkin I draw a diagram of an M10 full-nut. On the second napkin I draw a diagram of a 25mm M10 hex bolt. I send the first napkin to my father in England and ask him to make what's drawn in the diagram and mail it to me. I send the second napkin to the president of Honda in Japan and ask him to manufacture the bolt in the diagram and send it back. When the two parts arrive, they will fit together. Amazing eh? The reason I know that the parts will fit together is that there are international standards for the drawings, the manufacturing processes and most important, there are ways to test that the manufactured part conforms to the specification in the diagram. That specification not only defines the shape and size of the part but also the material it should be made from, the surface roughness, the hardness of the final surface and so on. When software is written the way nuts and bolts are made, software developers will have earned the right to call themselves engineers. So next time you fly to a software conference, be happy that Boeing doesn't build its planes the way Microsoft writes its software. ![]() |
||
|
|
|