As long as you insist on talking about the frequency of the test, people
are going to be ignoring you. Please start talking about what is wrong
with the test itself. Fixing the test will fix the problem. No amount of
tinkering with the frequency of the test is going to fix the problem.
The project came and went already. It was doing document indexing. The
runtimes were very short, but the transfer times were killing it.
jm7
"Paul D. Buck"
<p.d.b...@comcast
.net> To
Sent by: [email protected]
boinc_dev-bounces cc
@ssl.berkeley.edu TarotApprentice
<[email protected]>, BOINC
dev <[email protected]>,
04/27/2009 12:58 [email protected]
PM Subject
Re: [boinc_dev] 6.6.20 and work
scheduling
On Apr 27, 2009, at 5:47 AM, [email protected] wrote:
> Deadlines can be SHORTER THAN AN HOUR. There have been projects with
> deadlines of 5 MINUTES.
>
> I know I have stated this before.
And I know that I have stated this before. Sending tasks out to a
uncontrolled system like BOINC with deadlines of this nature is
insanity.
Likewise, it is insanity to drive all policy with an outlier project
like this if we also don't allow and controls to policy for other
exceptions.
Like, don't interrupt work in progress unless you have an emergency
like this ... and only then ... not for made up emergencies ...
Worse, most of the exceptions that are dragged out as examples lead to
rules that cause the systems to behave in very bad ways. I have 7 VTU
tasks running right now, all with 10 more minutes to run 56% complete
and a deadline of the 29th, meaning that this precious 5 minute
deadline maker rule is causing BOINC to over react and run tasks in
parallel where there is no need. Even granting the need, these tasks
have plenty of time to run in serial mode and none at all to have
started off in high priority mode.
As far as the frequency, ok, fine 5 minute task deadlines, the tasks
better damn well be 2 bytes long with a 10 second run time because
there is no other way that a task like that can be guarenteed a 5
minute turn around even on my fastest system with download, run,
upload if it has any sort of real work to do.
And by the way, which project has these tasks? I have never seen it
yet ...
The problem with accepting the argument that using event driven logic,
is that if the frequency of the events gets high enough the system
will become unstable. In the hypothetical 5 minute deadline I can buy
scheduling the task... if it is 2 days in the future, no ... the
problem is that there is no distinction... and the event driven model
means that a higher and higher percentage of time is going to be taken
up by this frenetic activity. And because BOINC recalculates
everything fresh it is always going to change its mind ... and the
faster the systems get with more processing elements the frequency is
going to keep climbing.
A dual Xeon with 16 CPUs and 3 GTX 295 cards is a total of 22
processing units ... using this event based model I can easily believe
that the frequency will be down to about once per second ... so which
project has a deadline 5 seconds in the future?
Well, my VTU shock wave passed and sure enough, BOINC started one new
task out of all that ... so a task that BOINC considered vital to run
a little while ago is now going to languish like a dead mongoose while
I wait for it to worry about it again ...
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
