Man of Honour
Running BOINC/VirtualBox on an SSD?
If you're running BOINC projects that involve virtual machines - such as vLHC@Home and ATLAS - on an SSD, and especially if you're running lots of them in parallel (I run 8), then I strongly suggest you consider the impact carefully.
Turns out that these projects eat gigabytes for breakfast, especially if you leave BOINC at its default checkpoint interval.
Even with an increased checkpoint interval (5 minutes), I chewed through nearly 6 TB in about 8 days. That's about 120 days worth of the 'average' lifespan that most sites use for SSD lifespan calculations. At that rate, a typical consumer SSD could be done in about two years, maybe much less for smaller/older SSDs. At the default checkpoint interval (1 minute), you could be looking at months.
If you're fine with that, then go right ahead in the knowledge that you'll be getting the best possible performance for your money. If you're not, then you may want to move BOINC data onto spinning platters. The easiest way to do this is to finish/abort all VirtualBox-related tasks (non-VirtualBox tasks will usually survive a move), uninstall BOINC, move the data, then reinstall BOINC and tell it where your data now resides.
Obviously, you'll lose some workunit performance by making this move, so as long as you don't reboot/crash the PC often, increase the checkpoint interval to at least 5 minutes, preferably more, to compensate, as that's where the majority of I/O happens. Also set BOINC to leave applications in memory when suspended if you can, otherwise you'll lose work back to the last checkpoint every time BOINC suspends.
Most non-virtual projects, with the possible exception of big data projects like ClimatePrediction, should be fine on an SSD, but again you should consider increasing the checkpoint interval and directing BOINC to leave tasks in memory to reduce wear.
All of these preferences can be found within the Your account page on any relevant BOINC project. Look for 'Computing preferences'.
If you're running BOINC projects that involve virtual machines - such as vLHC@Home and ATLAS - on an SSD, and especially if you're running lots of them in parallel (I run 8), then I strongly suggest you consider the impact carefully.
Turns out that these projects eat gigabytes for breakfast, especially if you leave BOINC at its default checkpoint interval.
Even with an increased checkpoint interval (5 minutes), I chewed through nearly 6 TB in about 8 days. That's about 120 days worth of the 'average' lifespan that most sites use for SSD lifespan calculations. At that rate, a typical consumer SSD could be done in about two years, maybe much less for smaller/older SSDs. At the default checkpoint interval (1 minute), you could be looking at months.
If you're fine with that, then go right ahead in the knowledge that you'll be getting the best possible performance for your money. If you're not, then you may want to move BOINC data onto spinning platters. The easiest way to do this is to finish/abort all VirtualBox-related tasks (non-VirtualBox tasks will usually survive a move), uninstall BOINC, move the data, then reinstall BOINC and tell it where your data now resides.
Obviously, you'll lose some workunit performance by making this move, so as long as you don't reboot/crash the PC often, increase the checkpoint interval to at least 5 minutes, preferably more, to compensate, as that's where the majority of I/O happens. Also set BOINC to leave applications in memory when suspended if you can, otherwise you'll lose work back to the last checkpoint every time BOINC suspends.
Most non-virtual projects, with the possible exception of big data projects like ClimatePrediction, should be fine on an SSD, but again you should consider increasing the checkpoint interval and directing BOINC to leave tasks in memory to reduce wear.
All of these preferences can be found within the Your account page on any relevant BOINC project. Look for 'Computing preferences'.
Last edited:
