From:	MERC::"uunet!CRVAX.SRI.COM!RELAY-INFO-VAX"  8-JUN-1992 09:46:28.23
To:	INFO-VAX@KL.SRI.COM
CC:	
Subj:	re: Summing AUTOGEN FEEDBACK files

	We have 6 Vaxstation 3100s that due to physical security regulations
	are only  allowed to be up 8 hours a day (9-5).  This is not long
	enough to get decent  feedback data.  If you look inside
	AGEN$FEEDBACK.DAT it is just a text file  and it looks feasible to
	merge these files together.  For each parameter one  would have to sum
	them, take the max or take the last.  DEC is appreciative of  my
	problem and has already considered it for a future release but I would
	rather not wait an indefinite period of time.  So has anybody out on
	the net  done any work with these AGEN$FEEDBACK.DAT files.  Some
	parameters it is obvious what is being measured but some I would like
	some more background on.

I've actually done this kind of thing by hand, but before I answer let me
raise another question:  Why do you think an 8 hour feedback set is not
sufficient?  The two main reasons one wants long sample periods are (a) most
systems have different loads at different times of day, different days of the
week, and so on, and one would like to make sure all of them are included in
the covered period; (b) some values are "peak" measurements and may take a
long time to grow to their actual maximum value.  Here, (b) is irrelevant:
Since the systems are NEVER up for more than 8 hours, no growth beyond 8
hours can ever matter.  (a) can only be relevant if your days vary signifi-
cantly.  If the variation is that on some days the machines are hardly used
while on others they are heavily used, then obviously you can just do an
AUTOGEN run using feedback from a busy day.  If you have distinct kinds of
busy days, just running AUTOGEN with each kind of feedback file will give
you an effect that's probably somewhere close to what a combined file would
have given you.

Anyhow:  I found it fairly obvious how to handle for various parameters.
Let's look at some samples from a V5.2-1 system:

	ACP_DINDXCACHE_CUR = 13
	ACP_DIRCACHE_CUR = 54
	ACP_EXTCACHE_CUR = 64
Various ACP-related parameters.  All _CUR values are the current values on
the system.  These should be the same in all the feedback files.
	DINDXATTEMPTS = 227947
	DINDXHITS = 219958
	DIRATTEMPTS = 161429
	DIRHITS = 155154
These are counts.  Sum them.
	FEEDBACK_NODE = "JERRY	 "
	FEEDBACK_SID = 134217728
These identify the system.  They shgould be the same in all the files.
	FEEDBACK_TIME = " 16-JUL-1991 13:59:06.18"
	FEEDBACK_UPTIME = 1346843
Obvious.  BTW, the UPDTIME is in seconds.
	GBLPAGES_CUR = 13700
	GBLPAGES_PEAK = 10862
	GBLSECTIONS_CUR = 256
	GBLSECTIONS_PEAK = 191
Current (maximum) and peak usage values for things that can be demanded during
system execution.  The _CUR values, again, should all be the same; you should
use the maximum _PEAK value.
	IRP_ALLOCFAIL = 0
Allocation failures.  Sum these.
	LOCKIDTBL_CUR = 355
	LOCKS_INUSE = 473
Again, a data structure size and the number of elements in use.  Maximize
the _INUSE values.
	MSCP_FRAG_IO = 0
	MSCP_TOTAL_IO = 100906
	MSCP_WAIT_IO = 0
Counts.  Sum them.
	NPAGEDYN_ALLOCFAIL = 0
	NPAGEDYN_ALLOCFAILPAGES = 0
More allocation failures; sum them.
	NPAGEDYN_REQUESTS = 1569692
Total requests, again to be summed.
	PAGEFILE1_NAME = "DISK$JERRY_54274:[SYS1.SYSEXE]PAGEFILE.SYS;1"
Should be the same.
	PAGEFILE1_PEAK = 8663
	PROCESSES_PEAK = 29
Maximum usages - take the maximum.

I found that there were at most 10 or 12 values that needed to actually be
combined, and the method of combination was always "sum" or "take max".
It's not hard to do by hand, and for any given version of VMS, it would be
a fixed process that you could easily automate if you had to do it more than
a couple of times.
							-- Jerry