Posts by Jesse Viviano

61) Questions and Answers : Wish list : Please make the VM jobs self-contained (Message 27907) Posted 22 Nov 2016 by Jesse Viviano Post: Please make virtual machine jobs self-contained instead of having to connect to the Internet to retrieve job parameters. My family's microwave oven makes my Wi-Fi connection unreliable. I have tried using the 5.8 GHz band to avoid the microwave oven, but the walls in my house absorb that range too much for that band to be of any use between my computer and where the network router with the Wi-Fi access point is located, so I am forced to use the 2.4 GHz band which the microwave oven wrecks when it is in use. Each time a virtual machine job starts from the beginnning and the microwave oven is running at that time, I get a computation error because the VM could not get the job parameters from HTCondor. BOINC handles unreliable connections much more gracefully than your current VM applications, and I would like that grace to be added to your project.
62) Questions and Answers : Wish list : Manual CPU code path selection in preferences (Message 27844) Posted 14 Nov 2016 by Jesse Viviano Post: I would like manual CPU application selection in our BOINC preferences. My CPU, an Intel Xeon E5-2690 v3 that I bought mainly because I wanted ECC memory support to help mitigate rowhammer memory security attacks, would run SSE3/PNI code the fastest out of the choices you offer because SSE3 and PNI are the same thing. However, I sometimes would get SSE2 or x87 variants of the application. First, x87 is deprecated on 64-bit mode and really only should be used if you need 80-bit precision because its stack-based architecture is really slow especially if the stack overflows, causing compilers to generate code whose first priority is to prevent stack overflows at all costs even if it costs speed. Second, SSE2 is much faster than x87, but SSE3 is faster still. (On a side note, AVX and AVX2 can really speed up your code on Intel processors with those extensions. On AMD Bulldozer family processors, many of them, especially Bulldozer and Piledriver, are slowed down with most AVX or AVX2 code, so you are better off writing SSE3 code with FMA4 support for those processors. Steamroller and later might have solved those problems so AVX2 might be the optimal code for them.)
63) Message boards : Number crunching : Wrong applications sent to my computer? (Message 27162) Posted 14 Feb 2015 by Jesse Viviano Post: Using that logic, the estimated gigaflops amount can be thrown off by tasks which terminate early (e.g. the simulated particles immediately crashed into the simulated walls of the LHC so there is no more need to track them). The first two runs that were returned from the no optimization application exited early. The third one that my computer returned went the full way and did not exit early. I guess the only solution without uninstalling and reinstalling BOINC to get a new computer number is to let the slow tasks process until they drag the estimated gigaflops average down.
64) Message boards : Number crunching : Wrong applications sent to my computer? (Message 27160) Posted 14 Feb 2015 by Jesse Viviano Post: I think that the estimation of gigaflops must be off because SSE2 and SSE3 allows four single precision floating point operations per instruction or two double precision floating point operations per instruction, while the x87 FPU only allows one extended precision floating point operation per instruction.
65) Questions and Answers : Sixtrack : Are wrongly optimized applications being sent to the wrong computers? (Message 27156) Posted 14 Feb 2015 by Jesse Viviano Post: I am using a Core i7-980X running 64-bit Windows 7. I recently noticed that my computer started to crunch SSE2 work units when it supports instruction sets up to SSE3 (a.k.a. PNI), and that it used to exclusively use the PNI application. I then let them finish and drain out. I then reset the project to try to get sent the SSE3 or PNI applications to get back to full speed. I knew that SSE3 was a minor improvement over SSE2, but I wanted maximum throughput. I was then given the no optimization 64-bit application, which the task manager shows is running in 32-bit mode by showing a *32 suffix to the process name. This application is much slower because the x87 FPU (the standard FPU for 32-bit x86 processors) is very slow and is the reason AMD made SSE2 the standard FPU for 64-bit mode. Could someone please fix the scheduler to send the correct applications to the correct computers? I had originally posted this in the number crunching board because I was unaware that there was a forum for platform independent issues. Scheduler problems are generally platform-independent issues in my mind.
66) Message boards : Number crunching : Wrong applications sent to my computer? (Message 27154) Posted 14 Feb 2015 by Jesse Viviano Post: I am using a Core i7-980X running 64-bit Windows 7. I recently noticed that my computer started to crunch SSE2 work units when it supports instruction sets up to SSE3 (a.k.a. PNI), and that it used to exclusively use the PNI application. I then let them finish and drain out. I then reset the project to try to get sent the SSE3 or PNI applications to get back to full speed. I knew that SSE3 was a minor improvement over SSE2, but I wanted maximum throughput. I was then given the no optimization 64-bit application, which the task manager shows is running in 32-bit mode by showing a *32 suffix to the process name. This application is much slower because the x87 FPU (the standard FPU for 32-bit x86 processors) is very slow and is the reason AMD made SSE2 the standard FPU for 64-bit mode. Could someone please fix the scheduler to send the correct applications to the correct computers?
67) Message boards : Number crunching : Host messing up tons of results (Message 27119) Posted 31 Jan 2015 by Jesse Viviano Post: Could this be one of the reasons the upload server filled up? Normally, when work units are successfully validated, the invalid results should be marked as invalid so that the invalid results' files can be deleted. It seems that your validator fails to mark the invalid results. Results marked as validation inconclusive generally have to stay in storage so that they can be compared to other results so that they can be validated against the resends. Changing them to the invalid state clears them for deletion.
68) Message boards : Number crunching : Error reported by file upload server: Server is out of disk space !? (Message 27108) Posted 29 Jan 2015 by Jesse Viviano Post: Reporting work units won't help clear out the old files because the transitioner is down as of this writing. Therefore, the validator will not know that it needs to validate any files. The assimilator cannot copy good results into the database due to not knowing it needs to do its job. The file deleter cannot delete any files due to not knowing that there are files that need removal.
69) Message boards : Number crunching : Error reported by file upload server: Server is out of disk space !? (Message 27103) Posted 29 Jan 2015 by Jesse Viviano Post: My uploads just went through.
70) Message boards : Number crunching : Error reported by file upload server: Server is out of disk space !? (Message 27101) Posted 29 Jan 2015 by Jesse Viviano Post: Could the problem be that the server is out of inodes? (In case you are wondering, an inode is a data structure that contains data about one file like file attributes and pointers to file block locations or to other pointers that point to either file block locations or other pointers to file block locations as needed depending on the file's size.) Since we are still able to post messages to the message board (which is on the same computer as the upload server according to the server status page), I don't think that the disks are out of free disk space blocks. As for Einstein@home, its biggest problem was that its table of inodes got completely used, so its server started having to search the inode table to find a free inode to handle a file upload, with each search taking 8 seconds. Einstein@home's long term solution will be to reformat the crippled server with a newer version of XFS that allows free inodes to be tracked with a b-tree. While this creates overhead in consuming an inode to create a file or releasing an inode when deleting a file because the b-tree needs to be maintained, searching a b-tree for free inodes when there are no unused inodes left is much cheaper than searching the inode table row by row for a free inode.
71) Message boards : Number crunching : Avast antivirus - false positive (Message 27099) Posted 29 Jan 2015 by Jesse Viviano Post: Please report the false positive using these instructions on Avast's web site.
72) Message boards : Number crunching : Error reported by file upload server: Server is out of disk space !? (Message 27098) Posted 29 Jan 2015 by Jesse Viviano Post: The transitioner, file deleter, database purger, the assimilators, and the test work unit validator have gone down as of this writing.

Previous 20

LHC@home