Posts by Travis DJ

21) Message boards : Number crunching : Server problems (Message 13717) Posted 23 May 2006 by Travis DJ Post: ...and you're actually having success with BAM? I've been signed up for a week now and still don't have a single project listed nor a CPID in the account to speak of. There seem to be a deal of bugs Willy is working through and my account's caught on the "bug" side of things. But at least I can say I'm not experiencing the host issue you have.. there aren't many factors which would cause a BOINC Client to keep generating new host-CPID (as opposed to user-CPID, which is where my problem with BAM lies). <grumbles>
22) Message boards : Number crunching : Work to be done! (Message 13716) Posted 23 May 2006 by Travis DJ Post: One thing I wonder about the work cache size is just how many % of people use that method specifically to get as much work for a particular project as opposed to using it (as it was intended to) to give computers who aren't always online enough work between connects. In regard to LHC it has always seemed to me the people who set their work cache high are simply into this to earn the most credit & the higher stats that come with it and nothing more - but my impression could be way off. In defense of my thoughts, LHC@Home hasn't ever been a seriously time-sensitive app and for that matter the average turnaround time staticstic has never worked (some of my hosts are 0 days, another 7.something days, another is 236278 days and utterly unrealistic :). My honest opinion about the level of work (not) available at LHC has more to do with CERN really putting themselves behind the power of free distributed computing which BOINC offers than it does anything else such as reliability, security, or the somewhat unpredictable job completion time. Each of those criteria have solutions except the commitment by CERN to fully exploit this awesome resource they have with us. That's my two cents.. didn't realize I had this much to say :)
23) Message boards : Number crunching : can't find workunit (Message 13639) Posted 17 May 2006 by Travis DJ Post: There is definately a problem. When I click on any of the WUs in my pending list I get the following error as well: Unable To Handle Request Can't Find Workunit Here are a few for reference's sake: 1339808 1343430 1346886 There are 31 results stuck in my pending list. Happy bug-hunting!
24) Message boards : Number crunching : WU's not being recognised by BOINC (Message 13624) Posted 14 May 2006 by Travis DJ Post: My overclock is at 210mhz up from stock speed of 200mhz. It has been this way since I first built this system. I suspected as much- when people experience the kind of lockup you had, either bad hardware or overclocking is the most likely cause. As discussed in other threads here at LHC, sixtrack is very sensitive to overclocking and it just makes sense given your symptoms and recent attachment. As far as your WUs.. take MikeW's advice. Just delete them. It's not worth it to try and salvage it.
25) Message boards : Number crunching : WU's not being recognised by BOINC (Message 13604) Posted 13 May 2006 by Travis DJ Post: How do I re-start or re-initialise the wu's into boinc ? That's very complicated. The WUs are listed in client_state.xml along with some additional information (checksum, size, state). You'd have to manually enter that information in for each work unit. Problem is doing it accurately and generating the checksum (if it's not identical to the one which was issued the WU will be invalid for "tampering"). So the easy fix: Start -> Run -> CMD Type "net stop boinc" Go back to the LHC@home folder and delete the affected WU zipfiles. Back at the command prompt type "net start boinc" Type "exit" Check your Boinc Manager to make sure no WUs are still being reported. (You may have to exit and open the BM after stopping the service) Wish I had better news for you especially since work has run out. [edit] I just saw your other message in another thread.. you wern't overclocking were you? :)
26) Message boards : Number crunching : Whats wrong with pending credit? (Message 13603) Posted 13 May 2006 by Travis DJ Post: While this project is active, there isn't a permanent admin to take care of such issues and it's a very low priority concern at best.. this has been discussed to great length in other threads about the same thing. See how things go over this summer - Chrulle reminded us the summer students will be coming and there is a possibility someone might get assigned to LHC@Home to do some admin work. Nothing's concrete.. we'll just have to see what happens. As long as this project still is useful to the scientists I'll be attached here.
27) Message boards : Number crunching : Are we dead in the water? (Message 13602) Posted 13 May 2006 by Travis DJ Post: I'm not sure I see the reason why the "no new work from..." messages in the log are an issue. Are you looking for an option to not report messages in the log? Do you parse or save your logs? Personally the current system works.. and as they say here in America "If it ain't broke, don't fix it."
28) Message boards : Number crunching : O.K., I'm New! No Work?? (Message 13592) Posted 13 May 2006 by Travis DJ Post: It could be that your BOINC client is working off some long term debt on another project first. Without seeing your client_state.xml file I can't really tell. An idea is to go to your projects tab and suspend SETI and Einstein, then select LHC and Update to see what happens.
29) Message boards : Number crunching : BOINC 5.4.9 new recomended version released (Message 13584) Posted 12 May 2006 by Travis DJ Post: From Berkeley's website: # Changes since 5.4.8 * Changed the master url messages from errors to warnings. # Changes since 5.4.7 * Notify the firewall and security packages of the screensaver's need to communicate with the client when the manager launches to avoid screensaver lockups due to the firewall and security packages. * Fix the wizard when running on the Mac so that all the text can be displayed. # Changes since 5.4.6 * Use the correct localized formatting for dates, time, and numerical values when displaying messages. # Changes since 5.4.5 * Attach to project wizard fixed. * NTLM Authentication for HTTP Proxies disabled. # Changes since 5.4.4 * More performance enhancements. * Connection failures due to network issues do not cause a backoff event. Retries are set for one minute. * SOCKS5 authentication bug fixes. It should now work. * HTTP authentication bug fixes. Last round of performance enhancments broke http auth. * Better network communitcation tracing through flag in log_flags.xml. # Changes since 5.4.3 * Performance enhancements. * Improve dialogs on Linux so they fit well on the screen and have OK buttons. # Changes since 5.4.2 * Fixed a shutdown problem which was causing the manager to linger around even after the core client had shutdown. * Fixed a window flicker problem on application startup. # Changes since 5.4.1 * Fix a crash when running under Win9x and screen saver passwords are enabled. # Changes since 5.4.0 * Screensaver fixes when running on a machine without a graphics accelerator. * Account manager display issue. I had been test running 5.9 for a week without issues on this host. Also running in addition to LHC: CPDN, LHC Alpha, Rosetta. No problems at all - works perfectly in conjunction with Trend Micro Internet Security 2006, d/l away.
30) Message boards : Number crunching : Overclocking Failed (Message 13574) Posted 12 May 2006 by Travis DJ Post: Ahhh.. I read you now. You're sure right about CPDN being a "Hell's Kitchen" of sorts - I feel sorry for all the North Bridges on Intel-based dual core CPU systems. Bottleneck city.. :(
31) Message boards : Number crunching : LHC@home Alpha? (Message 13573) Posted 12 May 2006 by Travis DJ Post: Chrulle, We really do appreciate and respect the fact you take your personal time to check on these boards every so often especially since your departure. :) Thank you!
32) Message boards : Number crunching : LHC@home Alpha? (Message 13568) Posted 11 May 2006 by Travis DJ Post: I had forgotten I signed up to the Alpha project on Feb 1 2005.. but it's crunching work from November 2005. I was never able to successfully attach to the project at the time and apparently it's fixed - and I'm crunching there but I feel it's a very very neglected project. No news, no updates, no idea why there is 7 month old work, problems with the php code, some account bugs.. it's really not worth attaching to. I only have one host donating 4% of its time.
33) Message boards : Number crunching : sixtrack 4.67 (Message 13553) Posted 11 May 2006 by Travis DJ Post: 1) LHC doesn't always have work avaiable, it goes through times when there is a lot and time when weeks go by without work. 2) Sixtrack is "optimized" already. 3) Your 3h48m crunch time is "good" for a system of your calibre. 4) AMD's Cool 'n Quiet is a CPU-Throttling technology similiar to Intel's SpeedStep. It "slows" your processor speed when your computer is not in use to save energy and keep the CPU cooler. It's a good thing to use; it is not dangerous to turn it off. Happy Crunching!
34) Message boards : Number crunching : Overclocking Failed (Message 13533) Posted 8 May 2006 by Travis DJ Post: Son Goku-- To elaborate more about CPDN: It's actually more memory intensive, but it really depends on the size of the CPU L1/L2 cache. The hadsm/cm/clm programs do extensive table lookups - Windows' performance monitor will reflect a large difference between the total number of page lookups & faults between sixtrack and hadxxx. [edit, was thinking about 2 things at once and made no sense out of something simple] On my AthlonXP 3200+ there was around (this is off the top of my head, it's been a few months since I ran it last) a 1:6 ratio of lookups between sixtrack and hadxxx.
35) Message boards : Number crunching : Overclocking Failed (Message 13532) Posted 8 May 2006 by Travis DJ Post: Son Goku-- To elaborate more about CPDN: It's actually more memory intensive, but it really depends on the size of the CPU L1/L2 cache. The hadsm/cm/clm programs do extensive table lookups - Windows' performance monitor will reflect a large difference between page lookups & faults between sixtrack and hadxxx. HADSM seems to perform better on my Pentium-M 1.6 (2MB L2 Cache, DDR333) than the AthlonXP 3200+ (2.2GHz 512K L2 Cache, DDR400) despite the higher frequency of the AthlonXP. That's what I've noticed anyhow. :)
36) Message boards : Number crunching : Will there be new work (Message 13531) Posted 8 May 2006 by Travis DJ Post: Might I humbly suggest the use of this smiley ? I second the motion..
37) Message boards : Number crunching : Overclocking Failed (Message 13521) Posted 6 May 2006 by Travis DJ Post: I got a few things from your post: 1) Overclocking and LHC don't work well with each other unless you do extensive testing prior to actually running LHC@Home on an OC system. The floating point math that Sixtrack does is sensitive to every digit to the end of the 80-bit register. One error at anytime during will kill that WU. 2) I'm guessing you have an ASUS motherboard and an AMD processor.. most folk with ASUS, MSI, Biostar (and others) sport an routine which automatically selects the best memory setting given various settings in BIOS. It is possible for a system to set a DDR400 chip to DDR333 if the timings (i.e. 2.5-3-3-6 1T @ DDR333 vs. 3-4-4-8 2T @ DDR400, DDR333 in this case is faster) are better at the "slower" speed. It's this reason DDR400 is "faster" than some DDR2-800 chips. Frequency means nothing when the chips have to waste more cycles waiting before it can do something else. So the memory may not be "false" as you say, rather it works best at a "slower" setting - what BIOS probably means is that after you've been playing with "FSB" in Windows and errors result .. then you reboot .. BIOS knows better and changes its settings back to a more compatible default and gives you an error message. 3) If memory seems to be your problem if you're OC why don't you test it whlie you go and make sure it works to your expectations? Visit http://www.memtest.org/ and download the .iso image of Memtest86+. If you have a fully supported chipset you will be able to test your ram and alter the memory timings while it's testing so you can quickly find the limits of your memory & motherboard. Again, keep in mind that a slower frequency with tighter timings will get you better results than high frequency with loose timings. ..just my two cents
38) Message boards : Number crunching : sixtrack 4.67 (Message 13407) Posted 19 Apr 2006 by Travis DJ Post: This is not entirely true. My BOINC doesn't isntalled as a service. And the power profile of my box is "Always ON". CnQ is on leaded to greater cruch time anyway. It seems like achim.huber post is totally correct and CnQ just takes thread priority into account. Out of curiosity, which cpu driver is in use? The one MS provides in SP2 or the one from AMD's site? In the past (with AMD's driver) the fix was either 'always on' or CnQ in bios set to 'disabled'.
39) Message boards : Number crunching : sixtrack 4.67 (Message 13393) Posted 17 Apr 2006 by Travis DJ Post: I turned off Cool'n'Quiet driver. Now it crunches two times faster(not only for LHC@Home). I guess Cool'n'Quiet doesn't really work as expected. According to my understanding a box shouldn't be slower if Cool'n'Quiet is on AND the box is fully busy. This topic has been covered before. CnQ will idle the cpu when it isn't under USER initiated loads. When BOINC is installed and is allowed to run as a service, THAT is what makes the difference. So when the user is not producing a workload (i.e. the PC is idling) CnQ will reduce the CPU speed to the lowest speed to match user workload. BOINC will crunch, but at low speed - this is default behavior for all CnQ enabled systems. The only fixes are to either install BOINC to run only when a user is logged in, turn CnQ off in BIOS or easiest yet, change your power profile to "Always On" in Windows XP/2000.
40) Message boards : Number crunching : How to Use 100% CPU? (Message 12968) Posted 8 Mar 2006 by Travis DJ Post: A great piece of advice concerning Intel Dual Core CPUs: Not a single Intel CPU presently directly connects two cores together Intel Dualies talk from one core to the front side bus to the second core (and vice versa). So when you're running memory-intensive tasks -- especialy climateprediction.net -- the fsb gets especially clogged up with because it has to handle CPU, RAM, PCI, PCI-Ex/AGP, and South Bridge traffic. AMD AthlonX2 chips do a much much more efficient job sharing memory since the memory controller is in the CPU itself and it doesn't a FSB to deal with in the way Intel dualies do. For example. The Intel 955 chipset has a 8.5GB/sec interface to the CPU. That bandwidth gets eaten up immediately under even partial loads where both cores are accessing RAM as fast as it can get it. Even using DDR2-667 with 10.2GB/sec bandwith, you'll never see the true performance of the CPU or the RAM for that matter when under full demand. AMD's solution provides up to 14.4GB/sec to the system (8.0GB/sec HyperTransport + 6.4GB/sec RAM interface). Despite the memory interface is only 6.4GB/sec however both cores can address memory without having to communicate outside the CPU and clog the bandwidth to everything else. It's a much nicer and clean solution. So I'd recommend an X2 cpu for you until Intel can come up with a better solution, like an integrated memory controller .. and they're still years off .. somewhere around 2007-2008 before they will have it in products. It's probably more information than you asked for but there is a good reason why AMD has the best dual platform especially for memory bandwidth hungry applications. Oh, and DDR2-667 has about the same real-world performance as DDR-400. DDR2 at that speed only has lower electrical consumption.

Previous 20 · Next 20

LHC@home