[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Log file for CM
On Fri, 18 Sep 1998, David M. Malon wrote:
> >
> > Luis Bernardo wrote:
> > >
> > > I evaluated the total transfer rate as a function of the time during the 6
> > > hours the tests ran (around 23000 seconds). The total transfer rate is the
> > > sum of the individual transfer rates at any given time (approximately).
> > > A plot of the transfer rate is include in the .pdf file I attach. As one
> > > can see when things go "well" we get a transfer rate around 3 MB/sec but
> > > most of the time things don't go well...
> > >
> > > Luis
> >
> > Arie wrote: It looks more like 6 MB/sec at the peak.
> >
>
> The highs (and some nonzero lows) may be artifacts of the rate-summing
> method. I am not sure of the method that Luis is using.
I only have the time a stage was requested, the time it was completed and
the file size.
>
> Suppose Joe and Bob each issue a request for a different 10-MB file at
> time 0. Suppose HPSS immediately services Joe's request, and delivers his
> file in 10 seconds, while Bob's request simply waits. Next, Bob's file is
> delivered during the following 10 seconds. The system delivers data at the
> constant rate of 1 MB/second for 20 seconds, but the rate summing method
> makes things look different. Because Joe's request was issued at time 0 and
> completed at time 10, Joe sees a rate of 10 MB/10 seconds = 1 MB/second.
> Because Bobs's request was issued at time 0 and completed at time 20, Bob
> sees a rate of 10 MB/20 seconds = 0.5 MB/second. If you sum the rates, it
> looks as though the system is delivering 1.5 MB/second in the interval
> [0, 10) and 0.5 MB/second in the interval [10, 20). (You can only
> believe the integral of the rates.)
>
Yes, that's what I do. The important point is that the test took 6 hours
because most of the time no transfer occured and the system just waited.
Otherwise one hour would had been enough. I think the QM only asks two
files at a time and waits for them to be staged before requesting the next
ones.
Luis
--------------------------------------------------------------
Luis M. Bernardo <lmbernardo@lbl.gov> <www.lbl.gov/~bernardo>
Scientific Data Management Research & Development Group
Lawrence Berkeley National Laboratory
Berkeley, CA 94720