Cary, I'd like to place my pre-order on that book right now please.
Rachel --- Cary Millsap <[EMAIL PROTECTED]> wrote: > Dan, > > I think I can convince you of the contrary, but I doubt that I can do > it > solely via email. In our Clinics, where we spend three solid days > addressing of all of the reservations you've stated below (including > proof with examples of why "holistic" approaches are unreliable, and > why > we believe response time analysis is actually the fastest route to > root-cause identification). The goal of the Clinic is to send you > home > able to actually *do* the stuff you see us writing about... > > One of the reasons I'm so confident in Response Time analysis is that > it > seems to be the final convergence for performance optimization work > in > other industries, some of which have been evolving for over a > century. > For a really nice description of one such industry, read Eli > Goldratt's > "The Goal." > > What Response Time analysis gives you that ratios don't is a valid > priority-based task order. To take your cue, you're right: a slow > response time on a query *could* indicate any of literally hundreds > of > different root causes. But seeing a Response Time account of exactly > what that query did will show you exactly where that query spent its > time. Consequently, you'll see exactly what you need to repair. > However, > if you're not looking at those response time statistics, then the > only > method you're left with is to check each of the hundreds of things > that > *might* have caused the query to be slow. Ratios are supposed to > guide > you through that process, but they often mislead you (as Dave Ensor, > Graham Wood, Anjo Kolk, Jonathan Lewis, Connor McDonald, Steve Adams, > Gaja Vaidyanatha, Kirti Deshpande, John Kanagaraj, James Morle, > Mogens > Nørgaard, Jeff Holt, I, and several others keep demonstrating). > > We visit several sites each year who have suffered the same > performance > problem for months or years. They've checked everything they can > think > of, and they haven't found the problem root cause. It is extremely > rare > for the same people to require more than one hour to positively > identify > and begin the repair of the correct root cause, when presented with > Response Time data. It's mind boggling how difficult it is to > convince > people to do it that way for the first time. Once they try it, they > never go back. > > * * * > > Within the next few months, I hope to have completed my book project > ("Oracle Response Time Optimization" or some-such). The questions on > this list are exceptionally good inspiration for such a project, by > the > way. When the book is finished, if I still haven't convinced you, > then I > simply won't have done my job. > > > Cary Millsap > Hotsos Enterprises, Ltd. > http://www.hotsos.com > > Upcoming events: > - Hotsos Clinic, Oct 1-3 San Francisco, Oct 15-17 Dallas, Dec 9-11 > Honolulu > - 2003 Hotsos Symposium on Oracle® System Performance, Feb 9-12 > Dallas > - Next event: NCOAUG Training Day, Aug 16 Chicago > > > > -----Original Message----- > Sent: Friday, August 09, 2002 12:09 PM > To: Multiple recipients of list ORACLE-L > > <Putting on t-shirt with honkin' big red bullseye target on the front > and > back> > > Stream of conciousness/random thoughts... > > To focus exclusively on response time is to fall victim to the same > fallacy > as focusing exclusively on hit ratios. The fallacy is that there is > one > and > only one methodology to use for tuning/troubleshooting. In 5 years, > another > approach will supplant response time tuning and we will all scratch > our > heads and wonder why Anjo and Cary could not see the obvious. Just > kidding...you guys will probably be the first to say "While the other > approach was good, here is why the new approach is better with this > new > technology/knowledge". > > The best approach, which is often preached, is to use the holistic > approach. > I recall a paper from some years ago that discussed this approach. > Most > of > the > tuning books advocate such an approach, but we techies get bogged > down > in > the day to day operations and only read the chapters that are of > immediate > concern. > > A poor hit ratio (of any cache) may indicate a problem, even if the > response > time is adequate. How many systems never increase in size or number > of > users? > > Response time monitoring describes the symptom. It does not define > the > root > cause. A large number of waits for the database writer does not > conclusively > determine a single root cause. Rather, there are many possible causes > (from > the cache to processes to i/o subsystem). A slow response time on a > query > could indicate a malformed query, missing/present indexes, poor > storage > parameters, an invalid high water mark, excessive read consistent > transforms, i/o contention, etc. > > One reason why hit ratios are nice is that they are tangible and > measurable. > Response time satisfaction is not. What is 'slow' to one user is > 'fast' > to > another. > > Thoughts...<ducking the inevitable flaming emails> > > Dan Fink > > -----Original Message----- > Sent: Friday, August 09, 2002 4:18 AM > To: Multiple recipients of list ORACLE-L > > > Cary, > > I love to talk about the response time thing, but it shouldn't be a > lunch but a big dinner (will also settle for Tuborg and hotdogs ;-)). > > > I think it is acutally great to see this discussion on the BCHR. I > don't > > except people to jump to response time tuning directly and drop the > ratio tuning thing. However the response time tuning approach works > great also for management. They like to hear things like 85 percent > of > the end user response time is in the network, because the hit ratio > could be perfect in this example (but the problem is outside the > database and the hit ratio will not show that .......). > > Reponse time/Throughput tuning is not perfect yet. There are may be 1 > or > > 2 tools that look at the whole stack and correlate that information. > There are no large number of tuning books about it, so people end up > buying the wellknown BCH books. Eventually the whole thing will > change. > Even Oracle changed STATSPACK in 9.2 to include the CPU time in the > TOP5 > > wait events (now why did they do that ;-)) > > I have to mow the lawn, the wife complains about the response time > (it > takes weeks to get a reply from me ;-). Will hear from you guys > later, I > > am sure ...... > > Anjo. > > > Cary Millsap wrote: > > >Sure, I'd love to comment... > > > >1. If you can inexpensively cache your whole database working set in > >memory, there's nothing wrong with doing that *unless* you could > have > === message truncated === __________________________________________________ Do You Yahoo!? HotJobs - Search Thousands of New Jobs http://www.hotjobs.com -- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Rachel Carmichael INET: [EMAIL PROTECTED] Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051 San Diego, California -- Public Internet access / Mailing Lists -------------------------------------------------------------------- To REMOVE yourself from this mailing list, send an E-Mail message to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing).