How to measure performance?

[David Collier-Brown]

Peter J. Holzer  wrote:
| This is not enough. Many applications do not read and write files
| sequentially. 

	Yes, but that's probably too much to ask Ms. Vologdin
	to do on the first go-round... 

| his is evident for MS-Access (which reads and writes
| single records), 

	I'm on the record as recommending one never try to run
	a multi-user database against a remote filesystem: search
	for "free sample of crack cocaine" in the archives (;-))

| 	and also for MS-Word (which only ever loads the part
| you are currently editing), but also for MS-Excel which does load
and
| save a whole spread-sheet at once, but reads and writes the file in
some
| seemingly random order (which is much slower).

| There are test suites like the BapCo suite which run the actual
| applications (but we couldn't find much difference between a local
disc
| and one shared over 10 Mbit ethernet, so at least Bapco is probably
more
| CPU-bound).

	Actually, oplocks cause local caching of the files,
	so the non-sequential accesses are mostly to local disk.
	That's possibly what fools bapco.

	[Andrew has a newish load-generation suite, specifically for
	SMB: I expect folks will eventually start using and reporting
	based on it...]

	Simple models and simple benchmarks aren't good for comparison,
	but in the presence of oplocks, can do a good job solving simple
	customer problems.

--dave
-- 
David Collier-Brown,  | Always do right. This will gratify some people
185 Ellerslie Ave.,   | and astonish the rest.        -- Mark Twain
Willowdale, Ontario   | http://java.science.yorku.ca/~davecb
Work: (905) 415-2849 Home: (416) 223-8968 Email: davecb at canada.sun.com