Posts by Lloyd M.

21) Message boards : Number crunching : Does the project need an equipment/ donation drive? (Message 16906) Posted 26 Mar 2009 by Lloyd M. Post: We've had a bit of a discussion along these lines on another thread. We were under the impression that MW didn't have the server to itself. A decent (at least) 2U server can be had for cheap. For instance, I bought a server with 2X2.5 gHz P4 Xeons, 1GB memory, all the hot swap fans (there are two on-board spares), and two 36GB or so HDDs for $75 off ebay. I got two 1GB RAM sticks for about $40 including shipping, and was able to use the two 256MB sticks I took out in a different server. Anyway, I didn't have to pay shipping, because the seller is close to my house (which gives me an advantage in bidding - sales tax is a LOT less than shipping for something like this). Still, there is a seller in Nebraska which has similar deals, shipping included, though usually without drives or drive caddies (or sometimes RAM). In my experience, by the time one gets surplus computers, the drives have a lot of hours on them, and the drives aren't as dependable as they should be for something mission-critical like this. I've found that one can expect to pay somewhere on the order of $60-$100 each for new, 73GB, server grade, hot-swap drives. The caddies cost something like $10 (or less - sometimes a lot less) on ebay. It seems to me that the minimum configuration would be two drives configured as RAID 1. If you're outfitting a "new" server, the ideal minimal configuration would be two RAID 1 arrays (one for OS and the like, and the other for data). One can get 36GB drives for somewhat less than 73GB (though nowhere near 1/2 the price), and if all the drives are the same size, one could use one "cold" spare for both arrays. Another thing to look at is hot swap fans and PSUs. One thing I like about my HP-Compaq server is that all the fans are identical, and two on-board fans are redundant. So if the fan right by the CPU Heat Sink dies, you can just grab one of the redundant ones and switch it in, without powering down the server. Fans can usually be had on ebay for about $10. A "real" server will usually have at least two hot swap PSUs (sometimes 3). Again, with the usual careful shopping, these go for about $20. My guess is that having the database on a different set of spindles than the rest of the server application(s) may well increase the throughput of the application, probably by a considerable amount. This counts even if you don't need another server. I would still strongly urge getting at least two matching drives and using a redundant RAID array (I know that sounds like I'm repeating myself, and RAID 0 is strictly for speed - it has no redundancy).
22) Message boards : Number crunching : WU abuse (Message 16903) Posted 26 Mar 2009 by Lloyd M. Post: The Gas Giant wrote: I just love it when people can't argue with facts and get abusive instead! Yes, I have found that there is a certain sort that resorts to ad hominem attacks.
23) Message boards : Number crunching : How do I get this type of point production? (Message 16901) Posted 26 Mar 2009 by Lloyd M. Post: Chris S wrote: It would not surprise me if this project goes out of business very shortly due to the use of GPU crunching. Why would having a GPU app cause the project to "go out of business"?
24) Message boards : Number crunching : WU abuse (Message 16900) Posted 26 Mar 2009 by Lloyd M. Post: Re: Gas Giant No credit for anything that won't run on an 8088! None of this new-fangled 8086 nonsense! :) Or maybe even Z-80... Uhhh - the 8086 predates the 8088. The latter was created because RAM cost so blooming much that it was too costly to have a 16 bit memory bus in machines like the original IBM PC. And the Z80 was pretty much an 8080 with double the compliment of registers - so it could run anything an 8080 could, but you could write code specific to the Z80 to take advantage of all the extra registers. Interestingly enough, a contemporary machine (the Tandy 2000) which was MSDOS compatible but not fully PC Compatible, had the vastly-superior truly 16-bit 80186 processor, which even had some functions moved from microcode to hardware, which allowed them to run faster.
25) Message boards : Number crunching : Server Updates and Status (Message 16899) Posted 26 Mar 2009 by Lloyd M. Post: Travis wrote: The GPU version of milkyway is going to farm more complex calculations to the application (gradient/hessian/line search calculations) which will let them do more "local" searches. This should increase the WU size of the GPU application by 300-1000 times what they currently are, significantly decrease server load, and also reduce the network usage of these clients. Sounds like a fine (and very clever) idea!
26) Message boards : Number crunching : WU abuse (Message 16896) Posted 26 Mar 2009 by Lloyd M. Post: The Gas Giant wrote: Over the years I have had available to me various intel based platforms. I've always believed that the credit granted as the generational improvements occured have been in line what was expected when compared to the older platforms. But I've always wanted more, more, More, MORE MORE MORE I TELL YOU! That brings up another point - how do you establish parity between different generations of CPUs from the same manufacturer. Many apps take advantage of specialized instruction sets (SSE3, etc.) that older CPUs don't have, and sometimes the resultant performance increase is dramatic. What shall we do to enforce "parity"? Refuse to use these specialized instruction sets, which could result in doing more science, all in the interest of "fairness"? You want to give people running older/less powerful CPUs enough credit so it's worth their electricity to run it (I had a Celeron box that I finally decided wasn't productive enough to warrant the electricty it used and heat it generated). On the other hand, especially with optimizations, this is going to result in the latest CPUs generating a LOT of credit. The answer is you don't establish parity because it is impossible to do so. I guess I'm kind of dense, as I just don't see who this hurts or how it's a problem.
27) Message boards : Number crunching : WU abuse (Message 16892) Posted 26 Mar 2009 by Lloyd M. Post: Paul D. Buck wrote: I do argue for cross project parity and I do not buy the argument that it is impossible. Well, it is impossible if the "powers that be" don't even want to look at the issue. But there is no technical reason that we could not fix this. There is a VERY simple and fundamental technical reason that "we could not fix this". Different projects process at different speeds on different CPUs. Some projects run considerably faster on Intel Core2 CPUs (especially optimized apps that can take advantage of their large L2 caches). Some projects run much faster on AMD CPUs. So where's your baseline? What do you use as the basis for your "parity"? Yes, I know that SETI is the proverbial "500 pound gorilla" here, and even they seem to leave a little slack in this regard, because (I'm guessing) they know how impossible true parity would be to implement. And all this doesn't even include GPUs. What happens when they have both CUDA and ATI apps for this project? I would literally bet the ranch that, when they do, the WU throughput isn't even going to be close - one is going to flat blow the other one away (I won't predict which). Even if they were about even, what's to keep someone from building an I7 box (or even a dual I7 box) with three Tesla 1060s and a 395 in it? Or four 4870s? Well, what except for the fact that they would never have enough WUs to feed it LOL. But you get my point - such a box would have astronomical credits/hour, to the extent that it had work to do. Until you can provide at least a tenable theory as to how your vaunted "parity" could be implemented, how about you stop accusing some conspiratorial "powers that be" for the lack of it? [edit] Tell you what. If you can't solve the "problem" how about you: 1) Explain in detail just exactly what "cross project parity" is. That is, if we were to have it, how would we know? 2) Explain to a poor, old, ignorant credit whore like myself why the inequity you perceive - this "unfairness" - bothers you so much? Why do you consider it a problem to begin with? If people choose to donate their resources to projects for their own reasons, and one of those reasons (perhaps the ONLY reason) is credit, how is that a problem for you? Would penalizing that person (or even eliminating credit altogether - as has been suggested here) benefit you in any way? And I have questions for all of you "I'm only doing this for the Science" types: If you really are, and you don't care at all about credit, then why do you care how much (or how little) credit we credit whores receive? Why do you resort to tactics like a whole team threatening to quit a project because it grants "too much credit"? "Too much", according to whom? [/edit]
28) Message boards : Number crunching : WU abuse (Message 16891) Posted 26 Mar 2009 by Lloyd M. Post: John Galt 007 wrote: OK, maybe 500 is a bit low...1k per core would get you about 200k credit per day on an i7, or any 8 CPU box. UL1 is doing 193k RAC with the top host. That should keep his PC fed... Can you explain to me what interest is being served by imposing artificial limits?
29) Message boards : Number crunching : WU abuse (Message 16890) Posted 26 Mar 2009 by Lloyd M. Post: Paul D. Buck wrote: I complain because the credit system is flawed and the payment is inherently unfair ... It's at times like these that I wish I was a lot better at math, so I could prove it. Intuitively, I don't think "payment" of credits is likely to ever be "fair" any more than anything else can be in this world.
30) Message boards : Number crunching : WU abuse (Message 16888) Posted 26 Mar 2009 by Lloyd M. Post: Lord Tedric wrote: As far as Crunchers are concerned, most projects are about credits, that's why they complain about it! Here we go again....
31) Message boards : Number crunching : WU abuse (Message 16887) Posted 26 Mar 2009 by Lloyd M. Post: Brian Silvers wrote: I tend to doubt just a second hard disk will alleviate enough issues to get it to where people are not clamoring about not having work, but I could be wrong. Actually, I remember being taught that moving your LDAP data to another spindle could result in a large throughput increase in AD services on a domain controller. I see this situation as possibly at least partially analagous, with a couple of disparate processes beating one drive (or array) to death with seeks all over the platter. BTW, that almost certainly is a SCSI drive. As for people "not clamoring about not having work"; I don't hold out much hope of that ever happening.
32) Message boards : Number crunching : WU abuse (Message 16886) Posted 26 Mar 2009 by Lloyd M. Post: Brian Silvers wrote: I think what a lot of people have trouble conceptualizing is that there can be hundreds or even thousands of requests per second. If there are 500 tasks available and 300 requests for an average of 2 tasks per second, some are going to be told that there aren't any available, while others get their 2 that they requested. That's just simple math. caferace wrote: Brian, perhaps if you had a system and and a current BOINC client that would fall into the parameters you seem to enjoy theorizing about things might be far more clear to you. As it is, I can tell you that much faster machines than your AMD 3700+ have issues that are exponentially outside your experience with BOINC and MW. Oh, wise one, please enlighten us with the mighty power of the wisdom borne of your vastly superior hardware!
33) Message boards : Number crunching : WU abuse (Message 16885) Posted 26 Mar 2009 by Lloyd M. Post: KWSN imcrazynow wrote: From what I see the slower systems have no problems running out of work. I think it's fair to say "very few problems running out of work". It has happened recently, even on one of my slower boxes. I do think that there's definitely some correlation between how fast your machine can process a WU and how busy it's going to stay, in terms of real time.
34) Message boards : Number crunching : WU abuse (Message 16884) Posted 26 Mar 2009 by Lloyd M. Post: Debs wrote: I don't see what is wrong with a carefully thought out script. A script that is written to hammer the server more than it needs to is something else. I agree totally. Debs wrote: I run a script that controls a number of projects, and uploads completed workunits every 40 minutes Seems reasonable. My fastest box (which isn't all that fast) tends to fail to upload some of its completed WUs. I really need to spend some time upgrading the client, because that is probably the problem. The slower boxes rarely seem to run out of WUs for this project, unless there is basically none being sent out by the server for a while. I do see "stranded", complete WU's here and there, though that seems to occur more frequently with the Windows boxes. I found it "dry" this evening, with about 6 completed WUs just sitting there, and nothing else to work on because I had all my other projects on that box NNT. I went ahead and allowed tasks on some other projects, which is good, because it's been about an hour and I still don't have any new tasks. Debs wrote: If I knew how to detect whether a specific project is waiting to upload or has run out of work, or better still how to tell how many tasks are in my queue waiting to start, I would write the sriptto check every so often whether a project is running short of work, and I would only connect at that time. I use boinc_cmd on my linux server box, which is strictly CLI. It displays all kinds of WU state values and whatnot. My guess is it's just reading some XML files somewhere - that's how this all seems to work. Debs wrote: And yes, I AM a credit "whore". I just don't have the hardware yet to reach the top 1000 here or in BOINC combined :) If only. I can't seem to break back into 5000 overall, though BOINC All Project Stats says I'm ranked 749th by TC and 515th based on RAC for this project. And, yes, I volunteer my CPU cycles to projects that I believe in, and collect credits (as intangible as they may be) in return. If that makes me a "credit whore", so be it. If I am, I think I'm in some very good company in that regard.
35) Message boards : Number crunching : WU abuse (Message 16882) Posted 26 Mar 2009 by Lloyd M. Post: Pwrguru wrote: The simple answer would be to just do away with all points....Then we would see how many people would still be here when the dust settles........ Yes, the "simple answer" would be to cut off our noses despite our faces, in complete ignorance of basic human nature, all for the sake of what is supposed to be a moral ideal, but is utterly unattainable in any case.
36) Message boards : Number crunching : WU abuse (Message 16881) Posted 26 Mar 2009 by Lloyd M. Post: Phil wrote: What - I can post a bigger logfile full of crap than you can!!! ROFLMAO! For my part, I was wondering what was going on with all of that.
37) Message boards : Number crunching : WU abuse (Message 16880) Posted 26 Mar 2009 by Lloyd M. Post: KWSN imcrazynow wrote: I currently have no ati card running MW. Feel free to check my computers. It is with actual cores alone. I can vouch for my fellow Knight. He has some serious hardware, and has had it since 'way before there was such a thing as a GPU app.
38) Message boards : Number crunching : Heart of Gold (Message 16684) Posted 23 Mar 2009 by Lloyd M. Post: Misfit wrote: Ahahaha! I'm still laughing at that question. Simple pleasures for simple minds.
39) Message boards : Number crunching : Heart of Gold (Message 16683) Posted 23 Mar 2009 by Lloyd M. Post: banditwolf wrote: Silver stakes in order? ...or maybe wodden bullets
40) Message boards : Number crunching : Heart of Gold (Message 16491) Posted 21 Mar 2009 by Lloyd M. Post: Lloyd wrote: who the heck accuses someone of "vampirism and lycanthropy"? Phil wrote: You'll read about it in various BIONC forums. Uhhh, OK. I'll take your word for it. I'm not going to spend the time looking for something like that That's what I like about KWSN - the forums are pretty much universally respectful, with studied silliness instead of the viciousness you see elsewhere.

Previous 20 · Next 20