[comp.sys.dec] BENCHMARKING VAXES

fuentes@kinks.rutgers.edu (Carlos 666 Trott) (04/06/91)

HI!

	I am currently having perfromance problems on several
vaxes we have.  Due to an unusual series of events we have 
outsourced them (we pay other companies to maintain them and keep
them at their site, but we get the entire machine).  Some of these machines
have had  severe performace problems, even when I am the only user
(Admittedly my programs are compute intensive).  At present, io & cpu
intensive programs run FASTER on my inhouse 3100 and on my 33MHZ 386
based PC.  (I am sure some system paramters are set wrong and am
investigating that possiblity).

	I am hoping that someone out there in Netland can help me out
by providing me with some benchmark programs that can be run on
different machines that are at different sites so I can compare them
and possible point out to the vendors which ones are improperly tuned.

	One example of one of my many dilemas is a 6410 (7 VUPS)
dedicated for our use at an outsorcing company.  At this site we 
are in a cluster with atleast 10 other VAXES.  This VAX is always having performace
problems and is being outpreformed by an 8810 (6 VUPS).  Identical
applications on the 8810 runs at over twice the speed.  (The 8810 has
an identical disk configuration, but is not clustered).

	Due to the nature of the environment the 6410 is in, we
cannot have privileges and are forced to rely heavily on ACLS.  They
are insisting this has minimal io perfromace degredation.  Is this
true?  Each file has 8 acls/aces.....and we use
identifiers rather than specific user names in the ACLs.   I don't
have any ACLS on our inhouse VAXES and there is a speed difference,
but I haven't measured it yet (hopefully this Sunday I'll get to it).
What kind of overhead do ACLS have?

	I was also wondering if there is anyway to measure if an HSC
is io bound.  From past expierence in setting up clusters I learned
that it was very important to "balance" the disk load as evenly as
possible accross HSC basically not putting the most IO intensive disks
on one hsc).  I'll use the example of the 6410 again:  we share the
same HSCs with other companies in the outsourced environment.  I know
they have very IO intensive databses and report generators always
running.  How can this affect our performance?  Is there anyway to
measure this?

	
	As is common is a clustered environemnt ther batch queue files
are shared.  What performance impact can this have on our batch machines or
other compaies batch machines in the same cluster?  I know batch jobs
sometimes take a little longer to startup...but are there any other
downfalls?

If you had to outsource some Vaxes, and Only needed to have one vax at
each outsourcing company.....would you prefer to be in a cluster (it
makes it easier for them to maintain and allows their clients to share
tapedrives)...or would you prefer to be standalone?  We really dont
need any of the benifits of a cluster...we don't need to share
disks... all we wish to have is shadowing which can be provided for by
a few HSCs.  What are the downfalls of being in a cluster when you
don't need to be?


I'ld appreciate any comments or benchmark programs (so I can compare
machines at different sites).

Please excuse me if my questions seem stupid, but I've worked in alot
of VAX shops and setup alot of small clusters, but I've never had
all of these problems.  Perhaps I'm overreacting because I don't have
control over the systems we have outsourced....but there are definite
differences between the outsourced systems and the in-house systems we
have.  The systems all run nearly the same applications, the only
difference is that the outsorced systems crank out a few hundred tapes
a week (at night) and our inhouse ones dont.  But the performance
problems occur even when we are not producing tapes.


-Carlos Fuentes, Jr.
Lead Systems Engineer
Citicorp*Lynch, Jones & Ryan
NY