Re: Why DO some people still prefer version 5 engines?
Posted in 1999
A discussion rather than a support case: why sites stayed on Informix 5 instead of 7. One poster cited a confidential benchmark where a tuned 7.x was ~20% slower on a three-table indexed join with ORDER BY DESC, and up to 20x slower in elapsed time when the query ran repeatedly in a stored procedure, despite lower CPU/system times; he couldn't publish details. Others argued 7.x wins on large databases and multiple CPUs, that poor or untrained tuning can make it slower, and that small 5.0 sites were fine until support ended. A user forced to upgrade reported MATCHES not using indexes (tied to NCHAR/NVARCHAR) and bad MAX results, acknowledged as an open, unfixed issue. No resolution is recorded.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
Topics: Performance & Tuning, Stored Procedures & SPL, Server Administration, Migration, Import/Export & Data Conversion, Cloud, Docker & Containers
Please tell us more about this unpublished research. Given that 7 is more tunable than 5, I can see how a malicious DBA could cause it to perform worse. For applications that are worth the money that 7 costs, it will almost always yield better performance than 5. Curiously, Clem At 05:51 PM 04-02-99 -0500, V. Lanny Rosicky wrote: >The reason is quite simple. There exist unpublish research which >clearly shows that Informix 5 is faster ander many cicumstances. >Sometimes (as in the case of strored procedures) 20 times faster. >But, oc course, if you have multiple physical cpu's or need the >functionality in 7, DON'T wait for 8 :-) >>On Wed, 3 Feb 1999, Andy Kent wrote: >> >>> Given that version 7 of SE and IDS are fundamentally superior to >version >>> 5, why are so many sites sticking with 5? And why do Informix >still go to >>> the trouble of upgrading them? >>> >>> Surely any compatibility problems have been resolved by now, and >the >>> migration path isn't THAT difficult. >>> >>> -Andy Kent- >>> Redland, Bristol >>> >>> >> > > > > _______________________________________________________________ Clem Akins (aka clem@informix.com) Menlo Park x5937 Customer Course Development Informix Software, Inc Last seen: Wallowing in the awsome winter splendor of Yosemite
Well, I would have already, if I was allowed to. Ask your Informix rep about a study done by Informix for IBM on the performance of RS/6000 single vs. multiprocessor boxes. We have conducted an extensive benchmark on our own apps. It had nothing to do with DBAs slanting the results, it was a _best_ tuned, Informix reviewed config of version 7. We concluded that on ordinary queries, both versions perfumed approximately the same. However, on a join of three tables, (join on single, unique index column) with a sort by desc, 5 outperformed 7 by about 20%. When we than, put this query in a stored procedure and ran it 10 times, back to back, 5 outperformed 7 multiple times. To be sure, the system and user times (sar) for 7 were impressively lower than 5 but the laps time was up to 20 times longer. For an OLTP application, the response time went from sub-second to several seconds and we found it unacceptable. Our new application does run on 7. We are avoiding those areas which were identified in our bench marks and also are using a 4 processor CPU. I certainly wouldn't want to try this app in 5, the functionality simply isn't there. I apologize for not being able to publish here something that I signed a non-disclosure agreement on - on the other hand, my 6 months frustration with Informix lack of explanation leads me to believe that I can, at least, post the fact that such a study exists. Lanny Clem Akins <clem@informix.com> wrote in message news:79dlvk$3tr$1@news.xmission.com... > >Please tell us more about this unpublished research. > >Given that 7 is more tunable than 5, I can see how a malicious >DBA could cause it to perform worse. For applications that are >worth the money that 7 costs, it will almost always yield >better performance than 5. > >Curiously, > >Clem > >At 05:51 PM 04-02-99 -0500, V. Lanny Rosicky wrote: >>The reason is quite simple. There exist unpublish research which >>clearly shows that Informix 5 is faster ander many cicumstances. >>Sometimes (as in the case of strored procedures) 20 times faster. >>But, oc course, if you have multiple physical cpu's or need the >>functionality in 7, DON'T wait for 8 :-) >>>On Wed, 3 Feb 1999, Andy Kent wrote: >>> >>>> Given that version 7 of SE and IDS are fundamentally superior to >>version >>>> 5, why are so many sites sticking with 5? And why do Informix >>still go to >>>> the trouble of upgrading them? >>>> >>>> Surely any compatibility problems have been resolved by now, and >>the >>>> migration path isn't THAT difficult. >>>> >>>> -Andy Kent- >>>> Redland, Bristol >>>> >>>> >>> >> >> >> >> >_______________________________________________________________ >Clem Akins (aka clem@informix.com) Menlo Park x5937 >Customer Course Development Informix Software, Inc >Last seen: Wallowing in the awsome winter splendor of Yosemite
Clem Akins wrote: > Please tell us more about this unpublished research. > > Given that 7 is more tunable than 5, I can see how a malicious > DBA could cause it to perform worse. For applications that are > worth the money that 7 costs, it will almost always yield > better performance than 5. Whoa, buddy, switch to decaf ;-) I think a key qualifier right here is "For applications that are worth the money that 7 costs". Surprise, surprise, there may be applications that aren't worth that money. (Actually, I suppose that depends on what you mean by "worth the money", because an application can be critical, without being a monster data-chewing resource hog.) Personally, I *like* 5.0. It works great on very little tuning; works better with some tuning and planning. If you have a small application/database that's running fine on 5.0, not outgrowing your machine, not giving any performance problems, and you don't want to invest a whole lot of time and energy into upgrading it, why not leave it on 5.0? Of course, 7.2/7.3 are great (and I don't just say that because people (used to) pay $250/hour to have me come and tune it for them ;-) If you need the extra features, want better performance, have a big database with huge tables, etc etc etc, switch to 7.x, (pay me lots of money ;-) and watch it scream. BTW, your DBA doesn't have to be malicious to make it perform worse, just untrained or incompetent. But I certainly wouldn't fault someone for staying on 5.0 if it's working for them. You must not have worked in Tech Support; you'd learn not to mess with success ;-) June -- june_t@hotmail.com Grounded in Palo Alto, living on M&M's (plain) Please do not send Informix questions to this account. I would add 'Please do not send spam to this account' but I suppose I would be wasting my bits.
On Fri, 5 Feb 1999, June Tong wrote: > If you have a small > application/database that's running fine on 5.0, not outgrowing your > machine, not giving any performance problems, and you don't want to > invest a whole lot of time and energy into upgrading it, why not leave > it on 5.0? That's been my position for several years, especially since I was told at several occasions by Informix employees that on our single-processor machine with only about 50 users max. and rather small databases (<< 2GB) there would be no benefit from switching to a newer release. Unfortunately, since Informix has discontinued support for 5.x and there is also no more support for 5.x from other software vendors like Legato, we were finally forced to make the upgrade. It went ok for about 98%, but the rest has been giving me lots of headaches (see my previous postings about Indices not being used with MATCHES clause and wrong results using the MAX aggregate function). > But I certainly wouldn't fault someone for staying on 5.0 if it's > working for them. You must not have worked in Tech Support; you'd learn > not to mess with success ;-) Never change a winning team. Too bad we were forced to. Regards, Richard -- +----------------------------+-----------------------------------------------+ | Dr. Richard Spitz | INTERNET: richard.spitz@lrz.uni-muenchen.de | | Weidenstr. 15 | Tel : +49-8141-33578 | | 82256 Fuerstenfeldbruck | FAX : +49-8141-346531 | | Germany | Mobile : +49-172-8933578 | +----------------------------+-----------------------------------------------+
In article <Pine.GSO.4.03.9902061612250.3738-100000@sun1.lrz- muenchen.de>, Richard Spitz <u7l11ac@mail.lrz-muenchen.de> writes >On Fri, 5 Feb 1999, June Tong wrote: > >> If you have a small >> application/database that's running fine on 5.0, not outgrowing your >> machine, not giving any performance problems, and you don't want to >> invest a whole lot of time and energy into upgrading it, why not leave >> it on 5.0? > >That's been my position for several years, especially since I was told >at several occasions by Informix employees that on our single-processor >machine with only about 50 users max. and rather small databases >(<< 2GB) there would be no benefit from switching to a newer release. > >Unfortunately, since Informix has discontinued support for 5.x and there >is also no more support for 5.x from other software vendors like Legato, >we were finally forced to make the upgrade. It went ok for about 98%, >but the rest has been giving me lots of headaches (see my previous >postings about Indices not being used with MATCHES clause and wrong >results using the MAX aggregate function). > >> But I certainly wouldn't fault someone for staying on 5.0 if it's >> working for them. You must not have worked in Tech Support; you'd learn >> not to mess with success ;-) > But why support 2 versions 5.x and 7.x surely it makes more sense once 7.x is relatively stable to just support 7.x. Less bugs to fix since you're not spending time fixing 5.x bugs. Also the people supporting 5.x can be moved to supporting 7.x hence the bugs in 7.x will be fixed faster. >Never change a winning team. Too bad we were forced to. > >Regards, Richard -- David Williams
Richard Spitz wrote: > Unfortunately, since Informix has discontinued support for 5.x and there > is also no more support for 5.x from other software vendors like Legato, > we were finally forced to make the upgrade. It went ok for about 98%, > but the rest has been giving me lots of headaches (see my previous > postings about Indices not being used with MATCHES clause and wrong > results using the MAX aggregate function). Ah, but that means you didn't simply upgrade. Your MATCHES issue is only a problem on NCHAR / NVARCHAR, which were not available under 5.0. So apparently you decided that while you were upgrading, you might as well take advantage of this new feature. If you'd stuck with CHAR, you wouldn't be having this problem. Not that I don't sympathize; I do. Just ask Snorri about our MATCHES / NCHAR problem. In fact, I entered that bug / "feature request" you are encountering (and spent a fair amount of time banging my head against R&D trying to get it "fixed"). I was going to reply to you about that issue, but I saw you found the bug yourself. Still, I am the "expert" on this "feature request", if you have any further questions, other than how to get Informix to fix it, which I obviously still haven't figured out. I'm thinking this might be a job for ... Superman -- I mean, IIUG. After all, IIUG is "International", and this is a problem for international users. What is needed is to band the affected users together to get the proper level of attention on this problem; each individual user that has encountered the problem has not had a loud enough voice. Too bad about the discontinued support for 5.0. I really liked that product. Nice little piece of software. June -- june_t@hotmail.com Grounded in Palo Alto, living on Hershey's Kisses Please do not send Informix questions to this account. I would add 'Please do not send spam to this account' but I suppose I would be wasting my bits.
So what kind of things do people trip up on to make 7.x run slower? -Andy- In article <79gdht$q27$1@news-1.news.gte.net>, june_t@hotmail.com (June Tong) wrote: > ... > > Of course, 7.2/7.3 are great (and I don't just say that because people > (used to) pay $250/hour to have me come and tune it for them ;-) If you > need the extra features, want better performance, have a big database > with huge tables, etc etc etc, switch to 7.x, (pay me lots of money ;-) > and watch it scream. BTW, your DBA doesn't have to be malicious to make > it perform worse, just untrained or incompetent. > > But I certainly wouldn't fault someone for staying on 5.0 if it's > working for them. You must not have worked in Tech Support; you'd learn > not to mess with success ;-) > > June > -- > june_t@hotmail.com > Grounded in Palo Alto, living on M&M's (plain) > > Please do not send Informix questions to this account. > I would add 'Please do not send spam to this account' > but I suppose I would be wasting my bits. > > >