10 MPoints in a day. WOWW

Absolutely marvellous. How to beat that monster cruncher.
Let's make some small ratios.
Mr. Kermit has registered 2600 devices to produce 10M points. This adds up to 3'846 points/device/day.

I use 6 devices at the moment and can produce on a good sunny day around 170'000 points. This means 28'333 points/device/day.
As all my machines are running 8 threads each (total 48 threads) that gives an average of 3'541 points/thread/day.

What is very interesting is that one of my threads equals one of his devices.

Flatly said to equal his crunching power I need 2'824 threads.
Let's see some paths of how to approach that monster. This is a theoretical exercise so we will not talk about the financial problem that will arise, the risks of divorce etc. etc.

By upgrading to 12 threads per CPU and upgrading all the devices in bi-processor I get 24 threads per device. That adds up to 144 threads.
In my future plan I will add another three devices. If this is done then I reach a grand total of 216 threads or 764'856 points/day. This is 13 times less then what I need.

So 764'856 points/day is the absolute maximum I can top with my planned number of devices crunching for WCG.

But an interesting question would be what is physical absolute maximum I can have in my very small basement room of which I am allowed 1/3 of the volume. After looking at the space available, cooling problems etc., it would be possible to fit 24 devices. I would then get a theoretical absolute physical maximum of 576 threads which would produce the marvellous quantity of 2'039'616 points/day.

As you can see, even that would not be enough, I will have to talk to my wife. I need some more space in the living room.

Daaarling .........

Hypernova,
The question is very basic: do you want and are you able to transform your own house in a computer center?
MrKermit has posted the details of what he can offer to WCG from time to time.
Also I think I remember you had problems with fuses once. So don't miss this line in the post just after the above one:
And forget about numbers of devices in WCG's stats, it is the least relevant item of these stats for various reasons.
Yesterday MrKermit has clocked 9:183:22:10:18 . So he is running about 3,469 threads. Approximately, because of WUs pending validation.

Cheers. Jean.

astrolab, I just looked at the number of registered devices that I can see in the member stats.
This means that he has 900 active computers but over time he has registered with WCG 2600 devices.
I have in my stats if you check 17 devices registered over a year. But only 6 computers crunching today.

Actually, we don't have near the number of devices as the "registration" says, because every time we re-join systems it doesn't recognize the identical name. We run RAM based root filesystems, so it's like an OS with Alzheimers, every reboot and it makes new friends.

This week we have ~560 Intel Dual Proc, Quad Core @ 2.8 Ghz and for a short while ~120 Intel Dual Proc, Dual Core @ 3.0 Ghz donating time. So on a Core by Core basis we have ~ 4,480 + 480 = 4960 cores crunching, with no Hyperthreading or oversubscription.

Cheers!
MrKermit

Hypernova,
MrKermit has shown that he wants to minimize transitional effects on the "WCG ecosystem" .
Also he has been in close contact with the techs when solving a teething problem that affacted his system.
So I am sure that he has taken all necessary measures to minimize the effects of an emergency stop of his contribution to WCG.

Running with a low work buffer is the key point to keep the number of potential "casualties" at a reasonable level.
Next, with help of BOINCCMD it is relatively easy to set all nodes to "No more work" for a smooth stop, or to detach them for an immediate cancellation of all active and queued jobs for an immediate redistribution.

Even if he did not care, as long as the work buffers have not been set to an extravagant number of days the effect on WCG operations would not be so catastrophic: there are between 60,000 and 100,000 WUs validated per day for each project, so the impact of a few thousands "Not responding" would be acceptable.

Edit: Added a link to the BOINCCMD wiki