Re: Oracle performance under NT

Stephen,
The problem was that each and every statement was being redescribed, reparsed, reexecuted which is highly CPU intensive. We would have seen the same behavior on any backend OS and hardware combination albeit to varying degrees. (NT,SUN,Netware, HPUX, AIX etc.)

Once we cleaned up our act in the application (used host variables propeperly - it is all well documented in the Oracle Manuals for application developers) things ran much better. Which is the point I ment to make below - times improved greatly with proper use of host variables from 22 minutes 100 % CPU usage to 6 minutes with 35% peaks with the same OS, hardware, test data and Oracle parameters (db block buffers, sga, etc., no disk reorgs)

We have been able to (with our application which is mainly an OLTP type) get up to 300 users on a quad Pentium Pro or about 75 - 100 users on a CPU (not strickly linear). These numbers allow response times to be within acceptable boundries (our judgement), and we try to be conservative in judging that. Your milage is going to vary from ours (plus or minus); since your needs and how you use the database will be different than ours is.

The only thing you can conclude in a broad sense - from the above - is that proper application design and use will gain you a lot regardless of the backend (NT, UNIX, etc.). Proper use of host variables in an OLTP environment will help greatly. Please don't assume that you can or cannot get 300 users on an NT machine. It just depends. What one means by X numbers of users is going to vary a lot and from the usually limited amount of information in these postings it is impossible to tell. (not enough characterization etc.)

Jim

Stephen Tenberg wrote in message ...
>This is a good example of an "urban legend" to me, i.e., someone's
>application stops working at 30 users, and they make the leap of logic to
>(NT + 30 Users = Real Bad).
>
>It is such a truism in development that the algorithm, plus the quality of
>the implementation, are usually more important than any tuning, or even
>often what product is selected.
>
>I have seen systems so poorly written that even a mainframe was not
>sufficient, successfully port to NT where they support the same number of
>users easily.
>
>It all comes down to the quality of implementation, IMO. Small platforms
>like NT just make gross coding practices easier to detect as there are less
>resources to waste.
>
>Steve
>
>
>Barbara Kennedy wrote in message ...
>>We experienced something similar to this. Funny the same server running
>>Oracle on Netware handled the task switching smoother and was less
>>noticible. We discovered that the application was not taking advantage of
>>host variables and thus causing every statement to be reparsed,
redescribed
>>etc. Once we fixed that the benchmark went from 22 minutes with 100% CPU
>>usage (flatline) on NT to 6 minutes with 35% CPU peaks - same machine,
same
>>data etc. Highly repeatable.
>>Jim
>>jacob bogers wrote in message <01be2b95$28b00ba0$17e6f1c3_at_msn01>...
>>>I had a project migrating an OTL (online transaction system) application
>>>from os2 TO WINDOWS nt4.0
>>>at ricoh espc in amsterdam holland. NT40 clogged up when about 30 people
>>>where
>>>accessing the database concurrently. (adding an extra processor, raid10
>>>controller,
>>>and 256meg (yes!) of memory to the machine didn't help at all. Not even a
>>>slight
>>>improvement. (I learned the hard way that NT40 isn't scalable.) After
>close
>>>examination it seems that NT40 stops when to many processes (28 to be
>>>exactly) are waiting for blocking IO. The kernel doesn't get any
processor
>>>time at all,
>>>(flat-line on the task manager monitor)!!
>>>
>>>/Jacob
>>>
>>>
>>>Thomas Pall <tpall_at_bga.com> schreef in artikel
>>><367847a9.0_at_feed1.realtime.net>...
>>>> It is entirely possible. However much will depend on how many indexes
>>>> have to be updated, amount of contention between users, how large your
>>>> SGN is, how fast and how many CPUs you have. You would certainly want
>to
>>>
>>>> go through the Oracle Server Tuning Manual and do everything you can to
>>>> speed things up.
>>>>
>>>> Michael Ho (hom_at_loyal-holding.com) wrote:
>>>>
>>>>
>>>> : Stephen Vear wrote:
>>>>
>>>> : > Does anybody have any performance statistics for Oracle running on
>NT
>>>? We
>>>> : > are considering a system that will have to perform about 40 table
>>>inserts
>>>> : > per second for a period every day. The inserts are about 6
>>>attributes each.
>>>> : > We basically want to know whether this sort of performance can be
>>>achieved
>>>> : > on an NT platform, or if we need to look to Unix.
>>>> : >
>>>> : > Any case histories are appreciated
>>>> : >
>>>> : > Stephen Vear
>>>> : > vears_at_logica.com
>>>>
>>>> : 40 table insert is okey for NT. Of course depend on your hardware
>>>setup,
>>>> : database design, database size, etc.
>>>>
>>>> : General Advice 1 : Never underestimate the system requirement, there
>>>are always
>>>> : some later come requirment cost 10 times of your resources.
>>>> : General Advice 2 : Never undersize your hardware, 'cause it is
>>>generally cheap
>>>> : right now.
>>>>
>>>>
>>
>>
>
>
Received on Sun Dec 20 1998 - 13:25:40 CST