Effect of disk striping on disk layout
All, My new company has an INFORMIX server that is agonizingly slow. Much of this may be due to the application running against it, but I can tell that disk I/O on this system was poorly implemented. I've suggested ways of balancing I/O (i.e. putting heavily hit tables on separate drives, and toward the physical center of those drives), the usual stuff. System Admin is more interested in disk striping to try to increase performance. I've told them that implementing RAID takes one form of performance tuning out of my hands. Based on your experience, would you recommend going with an improved disk layout, or with disk striping, or some sort of combination? Thanks very much for any insight you can contribute. Kind regards, -- <><><><><><><> John Bejarano San Mateo, CA <><><><><><><> --== Sent via Deja.com http://www.deja.com/ ==-- ---Share what you know. Learn what you don't.---
hey john it sounds like we work for the same company. I have a system that is io bound. what type of system hardware do you have to work with is the first place to look, and make sure you are not looking at the wrong cause. you can depending on your system do some simple things like move the page file to a different physical drive. this bought me about 8%. you can also move the log files to a different physical drive, this is my next step to be implemented soon. I will post the results. If you get any good info please pass it along. kirt Dearborn MI John Bejarano <jbejaran@my-dejanews.com> wrote in message news:7hhptl$j3n$1@nnrp1.deja.com... > All, > > My new company has an INFORMIX server that is agonizingly slow. Much of > this may be due to the application running against it, but I can tell > that disk I/O on this system was poorly implemented. I've suggested > ways of balancing I/O (i.e. putting heavily hit tables on separate > drives, and toward the physical center of those drives), the usual > stuff. > > System Admin is more interested in disk striping to try to increase > performance. I've told them that implementing RAID takes one form of > performance tuning out of my hands. Based on your experience, would you > recommend going with an improved disk layout, or with disk striping, or > some sort of combination? Thanks very much for any insight you can > contribute. > > Kind regards, > > -- > <><><><><><><> > John Bejarano > San Mateo, CA > <><><><><><><> > > > --== Sent via Deja.com http://www.deja.com/ ==-- > ---Share what you know. Learn what you don't.---
What version of IDS and which OS? John Bejarano wrote in message <7hhptl$j3n$1@nnrp1.deja.com>... >All, > >My new company has an INFORMIX server that is agonizingly slow. Much of >this may be due to the application running against it, but I can tell >that disk I/O on this system was poorly implemented. I've suggested >ways of balancing I/O (i.e. putting heavily hit tables on separate >drives, and toward the physical center of those drives), the usual >stuff. > >System Admin is more interested in disk striping to try to increase >performance. I've told them that implementing RAID takes one form of >performance tuning out of my hands. Based on your experience, would you >recommend going with an improved disk layout, or with disk striping, or >some sort of combination? Thanks very much for any insight you can >contribute. > >Kind regards, > >-- ><><><><><><><> >John Bejarano >San Mateo, CA ><><><><><><><> > > >--== Sent via Deja.com http://www.deja.com/ ==-- >---Share what you know. Learn what you don't.---
The system that I described in my previous post is IDS ver. 7.23.UC7 running on an HP K-series with HP/UX 10.20. I'm concerned about the I/O since checkpoints are taking 3-6 seconds for only 600-750K of buffer writes. That amount of transfer is barely more than 1% of my total buffers (which I had to increase because of the occurrence of ovbufs and foreground writes). Since it's such a small amount of info, tuning down LRU Max and Min to force LRU writes between checkpoints will do little if anything. All of this leads me to believe that disk layout is the issue here, or possibly even just *really slow* disk drives. Just this weekend we added a new tempdbspace on a different spindle which I hope will help, but that's just a temporary solution until a full re-org with a more intelligent disk layout is acheived. Kind regards, John. In article <7hn6qm$cc1$1@taliesin.netcom.net.uk>, "Neil Truby" <ntruby@netcomuk.co.uk> wrote: > What version of IDS and which OS? > > John Bejarano wrote in message <7hhptl$j3n$1@nnrp1.deja.com>... > >All, > > > >My new company has an INFORMIX server that is agonizingly slow. Much of > >this may be due to the application running against it, but I can tell > >that disk I/O on this system was poorly implemented. I've suggested > >ways of balancing I/O (i.e. putting heavily hit tables on separate > >drives, and toward the physical center of those drives), the usual > >stuff. > > > >System Admin is more interested in disk striping to try to increase > >performance. I've told them that implementing RAID takes one form of > >performance tuning out of my hands. Based on your experience, would you > >recommend going with an improved disk layout, or with disk striping, or > >some sort of combination? Thanks very much for any insight you can > >contribute. > > > >Kind regards, > > > >-- > ><><><><><><><> > >John Bejarano > >San Mateo, CA > ><><><><><><><> > > > > > >--== Sent via Deja.com http://www.deja.com/ ==-- > >---Share what you know. Learn what you don't.--- > > --== Sent via Deja.com http://www.deja.com/ ==-- ---Share what you know. Learn what you don't.---
Are you using kaio? (You can check this by running onstat -g ioq - if there
are kio vps listed then you are).
Neil
John I am a strong believer in the power of striping, and of RAID10 in particular, to improve the performance of Informix databases and take the burden of disk placement off my hands! Although I like Informix mirroring I strongly suggest that you NOT mirror stripe sets using either OS/firmware mirroring or Informix mirrors, either of which amount to RAID01 (or 0+1). Performance will be about the same as RAID10 but recovery is a nightmare and leaves you at risk for catastrophic data loss if ANY of the drives in the mirror stripe set should fail during or prior to the completion of recovery. Since RAID10 builds a stripe set from mirrored pairs only a single drive needs recover in the event of a drive failure and only the failure of that particular drives mirror can cause permanent data loss, an event make even more unlikely since it is many times faster to recover a single drive than an entire stripe set. RAID10 all the way! John Bejarano wrote: > > The system that I described in my previous post is IDS ver. 7.23.UC7 > running on an HP K-series with HP/UX 10.20. I'm concerned about the I/O > since checkpoints are taking 3-6 seconds for only 600-750K of buffer > writes. That amount of transfer is barely more than 1% of my total > buffers (which I had to increase because of the occurrence of ovbufs and > foreground writes). Since it's such a small amount of info, tuning down > LRU Max and Min to force LRU writes between checkpoints will do little > if anything. All of this leads me to believe that disk layout is the > issue here, or possibly even just *really slow* disk drives. Just this > weekend we added a new tempdbspace on a different spindle which I hope > will help, but that's just a temporary solution until a full re-org with > a more intelligent disk layout is acheived. > > Kind regards, > > John. > > In article <7hn6qm$cc1$1@taliesin.netcom.net.uk>, > "Neil Truby" <ntruby@netcomuk.co.uk> wrote: > > What version of IDS and which OS? > > > > John Bejarano wrote in message <7hhptl$j3n$1@nnrp1.deja.com>... > > >All, > > > > > >My new company has an INFORMIX server that is agonizingly slow. Much > of > > >this may be due to the application running against it, but I can tell > > >that disk I/O on this system was poorly implemented. I've suggested > > >ways of balancing I/O (i.e. putting heavily hit tables on separate > > >drives, and toward the physical center of those drives), the usual > > >stuff. > > > > > >System Admin is more interested in disk striping to try to increase > > >performance. I've told them that implementing RAID takes one form of > > >performance tuning out of my hands. Based on your experience, would > you > > >recommend going with an improved disk layout, or with disk striping, > or > > >some sort of combination? Thanks very much for any insight you can > > >contribute. > > > > > >Kind regards, > > > > > >-- > > ><><><><><><><> > > >John Bejarano > > >San Mateo, CA > > ><><><><><><><> > > > > > > > > >--== Sent via Deja.com http://www.deja.com/ ==-- > > >---Share what you know. Learn what you don't.--- > > > > > > --== Sent via Deja.com http://www.deja.com/ ==-- > ---Share what you know. Learn what you don't.---