Goodbye, BOINC...

I'm rapidly approaching the end of my year-long free trials with my various cloud instance providers. I've been running BOINC on them now all of this time, and now that I'll be pulling the plug I thought I'd write a post to properly "end" this topic on the blog. I have a tendency to let things "hang" so let's not do that.

If you were following along at home you'll remember that I had a frankly ridiculous 10 cloud servers. It may have even been 12 - I genuinely lost count. Two of those died fairly early on as I apparently burned through my Microsoft Azure credits quicker than anticipated. I thought I had them configured to last a year, but apparently not. So we were left with a smattering of Google servers and a couple of AWS servers.

One thing that did take some time to sort out was what projects to run on what hardware. It became clear that certain tasks would take too long, while others would be burned through very quickly. There was a higher score for doing those other projects, of course, so it's horses for courses. But I was very surprised at how different types of hardware seemed better at different types of calculation. I even saw massive variation in work unit completion between servers of similar types. I imagine this is down to some reason of configuration, but I couldn't see anything to account for it. Tens of percent in some cases, but that could be down to variation in the work units too I suppose. I didn't keep any real data along the way to measure this kind of thing.

Although I’m sure there have been some collection errors along the way I am sure I've made a bit of a contribution. I've done more data than 98.3% of all BOINC users, which is pretty cool. Through careful choosing of projects and configuration settings I was able to hold my own against people with vastly superior hardware. Every cloud server was a minimal install with only the extra bits for BOINC added. I do wonder if there was any more optimisation to be seen there, but I suspect it'd be minimal/not noticeable.

The good news is that the Google accounts do have smaller instances (f1-micro) that you can have for free forever, so I’ll be considering keeping those going but I don’t expect it will be troubling the stats too much. I’ll slip down the rankings but...

Decommissioning all of the servers was an exercise in memory recall - it’s been a while since I had logged into these servers so I kinda forgot all of the accounts I used. Took a while to test all the possible addresses I’d used. Heh. But eventually everything was cleaned up and I was looking at a very sad list of now dead servers in BOINC, all in red, as if to mark their passing.

Overall I am very glad I did this. It proved that it’s possible to play with the big boys when it comes to data crunching with minimal spend - for a while, at least. I have no idea what the monetary value of the resources I’ve used is. I probably should have checked. I've still kept up the Freecrunch project scripts and information, just in case anybody wants to do this. I genuinely think this CPU power is just going to waste. Anybody with a Google account can do this - and that's basically everyone, isn't it?