Message boards : Number crunching : How to upgrade to 64-bit without losing work in progress?
Author | Message |
---|---|
![]() Send message Joined: 16 Jul 06 Posts: 17 Credit: 1,083,385 RAC: 0 |
My question has nothing to do with 64-bit versus 32-bit but rather, I would be running a new executable and I don't want to wipe out the work in progress. I have four systems that all share the 400GB hard drive that is in one of the four. The other 3 diskless motherboards are booted off of USB thumb drives and use NFS as their own persistent storage. They have no keyboard, mouse, monitor, or their own drives, save for the bootable USB thumb drive. They are all linked together with a D-Link gigabit unmanaged switch. When I put 64-bit Fedora 11 on the newest one and overclocked it 20%, I noticed a marked improvement in performance as compared to one that runs 32-bit Fedora 10 also OC'ed 20%. So, I made up a couple more 4GB USB drives with that same 64-bit LiveOS on it, and want to run the other two diskless motherboards the same way. The BOINC installation of all four resides in seperate directories on the one shared hard drive. So my question is, if I change to a different version of BOINC, will it wipe out the work in progress? If so, is there a way of backing-up the work in progress? Can I just drop in a new executable and call it a day? I'm not trying to open the can-o-worms of 32 versus 64 bit. I don't give a $hit about that or the immature folks who act like babies when arguing the merits of one over the other. For my Q9550's and Asus motherboards, the 64-bit performs better. If it didn't, I wouldn't be going to it. That's all. Thanks in advance! |
![]() Send message Joined: 3 Nov 05 Posts: 1833 Credit: 120,108,280 RAC: 14,389 ![]() |
I think you can just install BOINC on your new image with the network share as the data folder and it should just run. It'll have to download the 64-bit mini-rosetta executable (you could do that in advance and drop it into the rosetta folder) but it's just the 32-bit file in a different wrapper. Probably worth backing the shares up just incase though... My question has nothing to do with 64-bit versus 32-bit but rather, I would be running a new executable and I don't want to wipe out the work in progress. |
mikey![]() Send message Joined: 5 Jan 06 Posts: 1896 Credit: 10,138,586 RAC: 20,966 ![]() |
My question has nothing to do with 64-bit versus 32-bit but rather, I would be running a new executable and I don't want to wipe out the work in progress. Widening the thought process here, mine not yours, so when you installed Boinc on each machines you just pointed it thru the network to your 400gb drive and it works that way? I have several Boinc only machines and that would be interesting to me. I have never used Fedora but Ubuntu is in the realm of possibility. USB drives are fairly cheap and I currently have a hard drive in each and every machine I have. That has always seemed pointless and a waste of my time to install an OS and then install Boinc just for a Boinc only machine. Now all of my machines are different models, specs, etc so cloning has never been a viable option. Hmmmm pondering the possibilities. Oh is the 400GB drive any where close to being full or could you add more machines to it? Did you have to make a directory for each machine and then install Boinc there? Or did you actually partition the drive into several partitions? As for the original question...I would run the cache dry then upgrade and then download new units. If you just set Boinc to no new work under the projects tab you will finish what you have and not get any new stuff. Then after the upgrade set it back to get new work and you will be off and running. You could probably even do it one machine at a time, or all at once, depending on your time schedule for doing the change. |
Mod.Sense Volunteer moderator Send message Joined: 22 Aug 06 Posts: 4018 Credit: 0 RAC: 0 |
I can confirm that when BOINC is installed and removed, it doesn't modify your data directory. But the rest will be up to you. Yes, just back it up first and if it somehow corrupts it then you can still complete that work on a 32b machine (any of them). Since the executable is specifically identified in the WU, I'm doubtful that you can just drop in a new one and have it run successfully against the existing WUs. But I look forward to hearing your results. Rosetta Moderator: Mod.Sense |
![]() Send message Joined: 16 Jul 06 Posts: 17 Credit: 1,083,385 RAC: 0 |
It has been working perfectly for a month now. On each diskless client, I add the following to the /etc/fstab: 192.168.10.103:/graviton /graviton nfs defaults 0 0 Where the 103 address is the one with the drive in it, and on it, there's a top-level directory named for each diskless client. Also locally, there's the same directory as a mount point. Now on the machine with the hard drive, in /etc/exports I have: /neutrino 192.168.10.101(rw,async) /nc6120 192.168.10.102(rw,async) /positron 192.168.10.105(rw,async) /graviton 192.168.10.108(rw,async) The 'async' option will keep your sanity. Without it, the hard drive will churn with every little bit of data that gets written over NFS. With async, the sending mahcine is assured that the data has been written, but in fact, it delays writing to disk until it is worth it to do so. You need to restart the nfs service with: "service nfs restart" Run 'iptables-save' on all machines and look for any default firewall rules that would prevent the free transfer of data between any two of them. Here's where it gets fun. After you install BOINC in each client's directory, you can run each client and have its display exported to the one with the hard drive. I have 4 "advanced" view clients off to the right of my dual monitors. Here's how to set that up: On the machine with the HDD, execute "xhost +" you'll be told that access control is disabled. Next, edit the 'run_manager' script for each client. Add the following line at the top of the script: export DISPLAY=192.168.10.103:0.0 Use the IP of the machine with the monitor/drive. Save, and exit. Now, when each client starts up, it will export its output to one machine. To make a LiveOS, make sure you have the livecd-tools package. I had to run "yum install livecd-tools" and from there, I plugged in a USB drive, found what /dev/ entry it is using the 'dmesg' command, then "umount /dev/sdb1". To make a LiveOS, it will insist that the drive be unmounted. The command I used to make the USB drive bootable is: livecd-iso-to-disk --overlay-size-mb 2047 Fedora-11-x86_64-Live.iso /dev/sdb1 then, wait about 20 minutes :) as it will initialize the overlay file. This will make changes you make to that filesystem, be persistent. With all 4 cores churning away, I've measured the power supply's consumption of this setup at 110 Watts. No monitor connected, no hard drive or floppy, no CDROM, just CPU+fan, motherboard, and memory, It's dead nuts silent, too. Hope this helps. Any questions, just ask. Widening the thought process here, mine not yours, so when you installed Boinc on each machines you just pointed it thru the network to your 400gb drive and it works that way? I have several Boinc only machines and that would be interesting to me. I have never used Fedora but Ubuntu is in the realm of possibility. USB drives are fairly cheap and I currently have a hard drive in each and every machine I have. That has always seemed pointless and a waste of my time to install an OS and then install Boinc just for a Boinc only machine. Now all of my machines are different models, specs, etc so cloning has never been a viable option. Hmmmm pondering the possibilities. Oh is the 400GB drive any where close to being full or could you add more machines to it? Did you have to make a directory for each machine and then install Boinc there? Or did you actually partition the drive into several partitions? |
mikey![]() Send message Joined: 5 Jan 06 Posts: 1896 Credit: 10,138,586 RAC: 20,966 ![]() |
It has been working perfectly for a month now. On each diskless client, I add the following to the /etc/fstab: That should do for now, THANKS! |
Message boards :
Number crunching :
How to upgrade to 64-bit without losing work in progress?
©2025 University of Washington
https://www.bakerlab.org