less v4 reservations being made

[tha]

Since about just over two weeks significant less reservations of exponents are being made by clients using the v4 server. The throughput of results turned in has not gone down. It therefore looks like as if the number of clients that reserve and discard exponents has gone down.

Whatever caused the sudden downfall is unknown, but it sure looks good.

[davieddy]

There must be a few months lag in converting reservations
into "throughput".
There seem to be plenty of discarded exponents on offer ATM.

[garo]

Guilty as charged. Team_Prime_Rib (i.e. me) discarded about 150 exponents in the 19M range that I had collected over the past couple of months.

[tha]

Originally Posted by garo

Guilty as charged. Team_Prime_Rib (i.e. me) discarded about 150 exponents in the 19M range that I had collected over the past couple of months.

150 exponents is a tiny drop of water in a basket full of it. Still, it accounts for a little of it.

I noticed the v4 server got remarkable more stable the last six weeks, cleaning up overdue assignments daily instead of irregularly, about once or twice a week. That might be the result of the v4 server not overloaded with assignments from one or more computers that do not do actual LL testing. That is just a guess.

Anyway, since June 18, 2008 no new assignments were made at above 43.700.000 which was the front line at that day. It looks like that will remain the case for the weeks to come. Before, the front line moved up about every week.

The result is that the work done is more focused in a smaller region of exponents.

[tha]

Originally Posted by tha

Whatever caused the sudden downfall is unknown, but it sure looks good.

about three days ago there was a sudden brief return to the old situation, many reservations made by many new accounts all named automatically by the Prime95 application. After about a day it stopped as well. That day the server did not do it's daily release of expired assignments as well, just like in the old days.

So it looks like there is only one client on the v4 server that is responsible for most of the reservations, and that client is blocking the v4 server as well. That client is not doing any real work. The only way to figure out more about this client would be to look into the IP addresses of the reservations, but that requires access to the v4 server.

[Old man PrimeNet]

For the last 3 or so days, here are the v4 server frequencies for IP addresses having 50 or more get-assignment requests. We could lock out the bandit if there's confidence it's not someone's legitimate workload.

134.*.*.* = 155
203.94.*.* = 150
4.58.*.* = 122
207.250.*.* = 107
128.187.*.* = 88
24.34.*.* = 80
80.64.*.* = 53
217.91.*.* = 50

[garo]

I'm 134.* and I'm legit. I usually try to harvest small exponents by reserving around 0600 UTC and then tossing back exponents that are of no use within a day or two. But 155 seems a bit high. I'll check my script.

PS: I don't think there is one runaway computer. I suspect tt's just lots and lots of people stress testing and reserving exponents they never meant to finish. Another check to run would be to see if a lot of the Sxxxxx accounts are coming from the same IP.

[tha]

Originally Posted by garo

PS: I don't think there is one runaway computer. I suspect tt's just lots and lots of people stress testing and reserving exponents they never meant to finish.

It is true that there are more computers of more users that make illicit reservations. However, the fact that all at a sudden the number of reservations drops by a huge percentage and than spikes back to the old level for just one day signals an runaway script or client. The more because in coincided with the day the server was unable to run it's daily clean up routine. Just like the old situation that suddenly ended 56 days ago.

Statistically it looks like about 300+ reservations extra were made in about 24+ hours. I have difficulties pinpointing the suspect reservations from the status.txt page, despite several analytical scripts I ran.

[ET_]

IIRC, older clients prior to version 21 suffered of that bug: they kept on reserving big chunks of exponents without completing them.

Luigi