Originally posted by: networkman
What I don't understand is why the Seti@Home project founders keep dragging this classic version on and on and on?
Well, the SETI@home/BOINC-servers haven't currently enough capasity to handle the load of "classic". Till they've moved the Master Science database from a E3500/6cpu with old, slow, nearly full non-raided hd's, to another E3500/8cpu with larger, faster and raided hd's, and configured the current server as data-server, they haven't the capasity.
This means, "classic" can crunch 1 day of recorded Arecibo-data per day, while SETI@home/BOINC will use atleast 2 days to crunch the same data.
If there's also a capasity-limit in the splitter-process so can at most split 1 day of recorded Arecibo-data per day, shutting down "classic" before SETI@home/BOINC is ready just means for every 2 days the backlog increases with 1 day...
Before got the new BOINC-db-server in February up and running, this taking so long among other things due to defective cpu, the capasity was so much lower so maybe needed 5 days to crunch one day of recorded data...
Anyway, let's say splitter-process is not a limiting factor, but there is another limit and this is bandwith. SETI@home's Cogent-link from campus is 100 Mbit, but to get to campus they must share the 100 Mbit SSL-link with the rest of SSL, meaning loses aorund 10 Mbit.
With every wu being 350 KB, this gives: 90 Mbit/s / (8 bit/byte) / 350 KB/wu * 86400 s/day = 2.777 M wu/day.
Yes.. One block is 1.7 seconds of telescope data. About 48 blocks are read in at a time (actually a bunch more are read in to "overlap" future workunits so there are no "edge effects" - we want to make sure if a Gaussian shows up it isn't split over two workunits and therefore scored lower or missed entirely). These blocks span the whole frequency spectrum, so we split the frequencies up into 256 equal parts, hence the 256 WU's per 48 blocks read.
And yes, following the math one tape yields about 176,000 WU's. Bear in mind these WU's are sent out redundantly.
So.. continuing along there are 86,400 seconds in a day. That's about 50,800 blocks. So we collect data to make about (50800/48*256) = about 270,000 WU's a day. At four times redundancy (max), that's 270,000*4 = 1,080,000 results a day that we need to handle. This is where our "million results a day" goal comes from.
- Matt (Lebofsky)
In other words, the internet-connection can handle roughly 2.5 days of recorded Arecibo-data per day, but due to BOINC client-downloads and seti-application-downloads and some results never returned and so on the effective capasity would probably be roughly 2 days of recorded Arecibo-data can be crunched per day.
This also means, as long as the capasity under BOINC is less than 500k "results"/day, shutting down "classic" N days before the capasity is there will take more than N days to catch-up. Not being sure at the launch 22.06.2004 how much hardware needed to add, nor if any serious bugs would surface when load increased, "cannibalizing" the "classic"-servers at the time could mean N days would need 10N days afterwards just to catch-up...
So, by this, not shutting down "classic" before the capasity really is available, makes sence.