No Subject

Below is a description of what we plan to log.

1)  For the tape drive activity, since we do not have information on
that, we'll log the "the number of PFTPs pending".  This can be
provided either by the CM, or by the QM (couting the number of files
requested but not satisfied at any point in time).  This number
should be logged whenever a caching request is made (i.e a PFTP is
issued) or whenever a caching request is satisfied (i.e. a PFTP
completed successfully).

Log (s_event): PFTP_PENDING
(for reference below, we call this value D - for Drive)

2)  For the cache activity, we should record how much of the cache is
actullly used by queries.  This can be done by generating the "total
of file sizes for files that are being used by some query".  This can
be obtained by the QM from "the set of files that are in use".
Dividing this by the cache size as set in the config file gives %
"% of cache in use".

Log (s_event): %_CACHE_IN_USE
(for reference below, we call this value C - for Cache)

3)  For the query activity, the QM needs to count 2 items: the
"total number of queries" in the system, and the "number of queries in a
processing status".  Both of these need to be logged.  In addition,
the ratio between them should be logged.

Log (s_event): NO_QUERIES_PROCESSING
Log (s_event): TOTAL_NO_QUERIES
(for reference below, we call the ratio of these values Q - for Queue)

Then, we can characterize the state of the system as follows: 

if  C > c and D <= d then --> SM_CACHE_STARVED
if  C <= c and D > d then --> SM_DRIVE_STARVED
if  C <= c and D <= d then -> SM_UNDERUTILIZED
if  C > c and D > d  and Q <= q then --> SM_WELL_USED
if  C > c and D > d and Q > q then --> SM_STARVED

where d,c,and q are constant that we select.
We'll use the values c = 80%, d = 10 and q = 50%. This information
will be logged whenever one of the values D,C,Q changes.

We plan to use a color scheme for the above measures, so that we can
observe the behavior of the system dynamically.

Comments are welcome.

Arie.