sellers@icni.UUCP (Dave Sellers) (10/31/90)
We have selected INGRES as the RDB (for many reasons which I don't need to get into). We also have VAXs as one of the main platforms, and have moved into SUN4 for workstations, and HP9000/870 machines in a smaller way. I have been asked about INGRES performance by a number of people, plus I have seen in this news group discussions about "hot" db machines (VAX not being one of them). I would like to pass along any info to others in the company who are about to purchase a DB machine. Did anyone keep any of this information? Is there any real applications running out there with 300 users? On what machine? Any measurements to base sizing on? Any rule of thumb? We have one project that has bought a VAX 9000 for INGRES without doing any benchmarks first! One comment so far is that "its the wrong machine for the job". Another comment was "if you have a choice between a fast single processor and slower multi processor box, pick the faster single processor box". Any comments? Please post here, or email via utzoo!censor!bert!icni!sellers Thanks -- Dave Sellers Data Communications Advisor Internal Communications Network - Implementation Bell Canada, 483 Bay St., Toronto, Ont., (416) 581-5913
lugnut@sequent.UUCP (Don Bolton) (11/15/90)
In article <747@icni.UUCP> sellers@icni.UUCP (Dave Sellers) writes: > >We have selected INGRES as the RDB (for many reasons which I don't >need to get into). > >We also have VAXs as one of the main platforms, and have moved >into SUN4 for workstations, and HP9000/870 machines in a smaller way. > >I have been asked about INGRES performance by a number of people, plus >I have seen in this news group discussions about "hot" db machines >(VAX not being one of them). > >I would like to pass along any info to others in the company who >are about to purchase a DB machine. > >Did anyone keep any of this information? Is there any real applications >running out there with 300 users? On what machine? Any measurements to >base sizing on? Any rule of thumb? > >We have one project that has bought a VAX 9000 for INGRES without doing >any benchmarks first! One comment so far is that "its the wrong machine >for the job". Another comment was "if you have a choice between a fast >single processor and slower multi processor box, pick the faster single >processor box". Any comments? > >Please post here, or email via utzoo!censor!bert!icni!sellers > >Thanks Though my employer would have a vested interest in my statements here, I speak for my own perspective... The multi processor "box" as you call it will benifit you in user throughput, even if your RDBMS has not been optimized for paralell processing. Think of the check-out lines at the super market. ASSUME 1 super fast clerk does 1 customer per minute. The typical clerk does 1 customer per every 2 minutes. EVALUATE The fast clerk can do 12 customers in 12 minutes 4 typical clerks can do 12 customers in 6 minutes CONCLUDE in "man hours" the fast clerk is super efficient in elapsed time the 4 typical clerks are twice as fast. Course this scenario only applies to a symmetrical multi-processing super market. ;-)
gnb@bby.oz.au (Gregory N. Bond) (11/16/90)
>>>>> On 14 Nov 90 22:28:47 GMT, lugnut@sequent.UUCP (Don Bolton) said:
Don> Though my employer would have a vested interest in my statements here,
Don> I speak for my own perspective...
Don> The multi processor "box" as you call it will benifit you in user
Don> throughput, even if your RDBMS has not been optimized for paralell
Don> processing.
Only if you have enough active users that there are more than one db
task active for most of the time. That is surprisingly rare!
Our situation, even though we had plenty of users, was that the
wallclock time for each operation was too high, and that it was
extremely rare to have two different backends active at once (we use
Ingres 5, with a backend per process). So a sequent-style machine was
not a good choice for us. We bought a single CPU Solbourne 5/600,
which had much better wallclock time for a single user, as well as
handling more users faster than our old server (a Sun 3/260).
A Wild Ass Guess would seem to indicate that 20 or more users running
applications more or less normally would be required before there was
a sufficient number of distinct backends in the run queue for a
multi-cpu to be a gain, at least with normal sorts of enquire/update
applications.
This sort of assumes a dedicated DB machine. If you were running front
ends, X, logins, compiles or other compute-intensive stuff on the same
machine, then the multi-cpu will become attractive much sooner. And
the Solbourne can take up to 8 processors if that is ever a problem.
Greg.
--
Gregory Bond, Burdett Buckeridge & Young Ltd, Melbourne, Australia
Internet: gnb@melba.bby.oz.au non-MX: gnb%melba.bby.oz@uunet.uu.net
Uucp: {uunet,pyramid,ubc-cs,ukc,mcvax,prlb2,nttlab...}!munnari!melba.bby.oz!gnb