....lasted 4 seconds

I got two more short ones after my first two.

They only lasted 5 1/2 mins, I have a feeling all the main beta units are gone and whats left are the child units required to complete the main units.

I've picked up a 3hr child one.

As a suggestion to any techs reading this, when the beta tests are sucessfully completed and you are ready to lengthen the target WU length, could you do it in stages?

I'm thinking that there will be people out there with 2 day plus buffers and if the WU's go from 1.5hrs to 8hrs in one leap Boinc won't adjust the average completion time until the first 8hr unit completes and by then it'll be too late as the buffer will now be full of 8hr units that Boinc assumed would be 1.5hrs. i.e. 8 days worth of WU's which Boinc previously thought were 1.5days. Even running in panic mode a buffer of any more than 1.5 days wouldn't clear before the units 10 day deadline. This would result in wasted crunching effort.

If the target WU length were raised by say 30mins per day when they are issued by the server the clients wouldn't get such a nasty shock and would have a chance to adjust their calls for more WU's.

A more elegant solution would be if the servers could in some way signal to the Boinc client to reset their averages for the longer WU's.

The beta is largely finished. There is only 1 workunits left.

The last set of workunits were interesting as they were children of some very tough workunits. Those tough workunits were interesting in that some positions were very easy (a few minutes to compute) or tough (requiring several hours to compute). Here is some details about the workunits:

Generation		Num Workunits	Avg Minutes	Stddev Minutes
Parent			10,000		136		70
Child			15,267		42		58
Grandchild		3452		21		60
Great-grandchild	1257		14		53
Great-great-grandchild	76		13		48

As you can see from the above this is not quite ideal. Our ideal is that avg minutes should remain constant throughout the descendants - instead it gets progressively shorter. We hope that having longer workunits the sample period will be better (i.e. multiple 'tough' positions can be computed within a single task instead of just one) and child generation will be correspondingly better. We will be watching this closely after we release this into production.

However, the critical item is that we have put a cap on the max duration of a task assigned to the members computers and we can release longer running workunits with this technique and know that we will be able to complete the tasks.

We are going to set max duration to 8 hours. However, that limit will only be checked after each position. On a slow computer you might, very infrequently and in the worst case scenario, hit a workunit that runs 10-12 hours.