Hi Gregory,

On Mon, Aug 30, 2004 at 01:51:57PM -0400, Gregory J. Sharp wrote:
> In the etc/ and etc/ example files we find the 
> following line.
> RESP=`/usr/local/bin/taylor -q opts | grep '^olbd=.*master'`
> 1. We don't have a program called "taylor" and I wondered what it does.
> 2. In the case of the, should it really be grepping for 
> oldb=.*master, or should it be looking at xrootd=.*master??
> As written it would seem to force the olbd master and the xrootd 
> redirector to run on the same host, which may well be efficient and 
> desirable, and therefore a "good thing", but perhaps not conducive to a 
> clear understanding of how things work (as evidenced by my suspicion 
> about why it is done this way).

  In the examples, I've basically ignored the startup scripts and those
example config files for the moment. (In part because they are SLAC-specific 
and somewhat confusing. "taylor" is a SLAC-specific system for distributing 
software to specific nodes.) Tomorrow I'll come back to finishing the
non-SLAC specific load balancing example, as well as finishing the SLAC
example with an explanation of these things.

  At least as we use the system there is no reason not to run the olbd
master and xrootd redirector on the same machine. See the SLAC example
to see the hardware we use:

The average file open rate isn't actually that large. From the xrootd
log files on the redirector I can see that it varies from 0.5Hz to 2Hz 
depending on the day (40-160 kFiles/day, say). Our data is relatively well
organized so a file open usually results in someone reading at least 100's
of MB (I think, we are adding something to log that, too) and it could be
up to 1 or 2GB per file open. There are certainly transient situations were
the rate is higher than that, but the hardware itself hasn't been a 
limitation. The redirector (with both redirector xrootd and manager obld)
are very lightweight and don't use significant resources.

  (Some one else familiar with "taylor" will have to answer the question
about as to what it should be grepping...)

  BTW, feel free to call into the weekly meeting if that is useful for you.
(Either to "lurk" or to ask questions.)


Peter Elmer     E-mail: [log in to unmask]      Phone: +41 (22) 767-4644
Address: CERN Division PPE, Bat. 32 2C-14, CH-1211 Geneva 23, Switzerland