[RC5] Block sizes (nkeys)

Hilmar Preusse hp4 at Rcs1.urz.tu-dresden.de
Mon Apr 6 19:38:06 EDT 1998


On Sun, 5 Apr 1998, waldo kitty wrote:

> OK, the way I understand it is this:  the first thread grabs a block and
> gets it ready.  Then, the second thread starts up and starts cracking
> another block.  When the second thread finished its block, the first hands
> its block to the second, and, while the second thread merrily continues
> cracking, the first grabs another block and gets it ready.  And so on...

And now what I see:
I'm running on a Solaris-machine with 4 CPU's.
problem:  This machine has an kill-client installed. This thing looks every
hour at the process-table and kills every process, which is longer than
half an hour in the CPU. Cause I'm running an 4-thread client in real time
this is for me 7.5 min. I have a script, which turns down the client and
up 90 sec. later. This script is started every hour (crontab) shortly before
the killer comes. It works very fine. If I would use the function of time
limiting of the client I'm afraid I would get trouble, cause the old client
stops at the same time, when the new client is turned up -- that's why the
delay of 90 sec.
The client takes 8 blocks in the CPU on every turn up and starts running on
only 4 of them. When one of these blocks is finished it takes one of the 4
blocks from it's 'master threads' and starts running on that. But when it
is turned up again the 'master thread' grabs the 'old' block, which was par-
tially done in the last run, and the 'child thread', which is really doing
the job grabs a new block, which was not done before. The following is,
that I have here a block where the client is only 50% through and I have no
idea, when it was fetched from the proxy. I'm afraid it can be here still
weeks, cause it's mostly in the 'master thread', which helds it for the
'child thread'. My estimation is, that it will be done at the end of the
week :-(
						Hilmar
--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list