Re: RE: Stripesize on Informix Raw devices

IBM own benchmarks recommend for disk throughput stripe size 64K max_coalesce 64 minpgahead 2 mapgahread 16x<no drives> Don't forget to alter maxfree Also if you are using 2 or more disk adapters you must use 64 bit aligned I/O buffers to avoid LVM serialization of the I/O. Paul Watson WF Software Ltd Tel: +44 1436 674729 Fax: +44 1436 678729 www.wfsoftware.com/informix # If you broke it, hide the evidence >>> "Bernstein, Rick" <rbernste@alarismed.com> 10/12/99 03:03AM >>> Does the recommendation of a 16K stripe size also hold for AIX platforms and others which have a 4K page size? Or should the stripe size by 32K (or 8 pages)? Your recommendations would be appreciated. Thanks, Rick Bernstein > -----Original Message----- > From: Art S. Kagel [mailto:kagel@bloomberg.net] > Sent: Monday, October 11, 1999 10:34 > To: informix-list@iiug.org > Subject: Re: Stripesize on Informix Raw devices > > > The best stripe size for Informix is 16K. Informix either > does BUFFER > writes of 2K or BIGWRITES of 16K so the 16K normally works out best. > Remember that Informix is not normally write intensive, due to it's > excellent buffer cache, so unless you are really stressing > the system it > may be very difficult to detect any improvement from striped -vs- > singleton drives. You will see the biggest improvement from watching > your checkpoint durations during load. Data load tests will > rarely be > affected by disk layout unless the loadset it MUCH larger > than the buffer > cache and is being inserted by MANY independent processes or threads. > > Art S. Kagel > > Marc Hopkins wrote: > > > > Does anyone have any experience with striping data on > Informix raw devices. > > We are running HPUX 10.20 on K420's. Informix reads and > writes in 2K pages. > > I have tried the stripesize of 4, which is the minimum HP > allows, and > > performance has not been any better than non-striped data. > I have the data > > striped across four disks. Anyone else using a diff > stripesize? I am not > > sure if I need to take into account the Informix Read-Ahead > configuration. > > Any suggestions welcome. Please send responses to email. > > > > Thanks, > > Marc > > marc.hopkins@aei.com > ********************************************************************** This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the system manager. This footnote also confirms that this email message has been swept by MIMEsweeper for the presence of computer viruses. www.mimesweeper.com **********************************************************************

Yes and DG recommends 128K for filesystems. HOWEVER, for Informix a smaller stripe size is essential to good performance EXCEPT for DSS/DW installations where read ahead is VERY heavy. Overall for most Informix systems, and all Informix OLTP installations, 16K (or 32K for 4K page systems) is best. Some vendors, notably Sun, Data General, and NCR have performed the benchmarks using Informix and most of the studies that I have seen reinforce this value. Don't trust filesystem benchmarks, they are heavily skewed by random I/O testing and while it goes against one's gut feel, Informix engines do not behave as random I/O requestors in production due to the excellent buffer cache. Art S. Kagel Paul watson wrote: > > IBM own benchmarks recommend for disk throughput > > stripe size 64K > max_coalesce 64 > minpgahead 2 > mapgahread 16x<no drives> Don't forget to alter maxfree > > Also if you are using 2 or more disk adapters you must use 64 bit aligned I/O buffers to > avoid LVM serialization of the I/O. > > Paul Watson > WF Software Ltd > Tel: +44 1436 674729 > Fax: +44 1436 678729 > www.wfsoftware.com/informix > # If you broke it, hide the evidence > > >>> "Bernstein, Rick" <rbernste@alarismed.com> 10/12/99 03:03AM >>> > Does the recommendation of a 16K stripe size also hold for > AIX platforms and others which have a 4K page size? > Or should the stripe size by 32K (or 8 pages)? > > Your recommendations would be appreciated. > Thanks, > Rick Bernstein > > > -----Original Message----- > > From: Art S. Kagel [mailto:kagel@bloomberg.net] > > Sent: Monday, October 11, 1999 10:34 > > To: informix-list@iiug.org > > Subject: Re: Stripesize on Informix Raw devices > > > > > > The best stripe size for Informix is 16K. Informix either > > does BUFFER > > writes of 2K or BIGWRITES of 16K so the 16K normally works out best. > > Remember that Informix is not normally write intensive, due to it's > > excellent buffer cache, so unless you are really stressing > > the system it > > may be very difficult to detect any improvement from striped -vs- > > singleton drives. You will see the biggest improvement from watching > > your checkpoint durations during load. Data load tests will > > rarely be > > affected by disk layout unless the loadset it MUCH larger > > than the buffer > > cache and is being inserted by MANY independent processes or threads. > > > > Art S. Kagel > > > > Marc Hopkins wrote: > > > > > > Does anyone have any experience with striping data on > > Informix raw devices. > > > We are running HPUX 10.20 on K420's. Informix reads and > > writes in 2K pages. > > > I have tried the stripesize of 4, which is the minimum HP > > allows, and > > > performance has not been any better than non-striped data. > > I have the data > > > striped across four disks. Anyone else using a diff > > stripesize? I am not > > > sure if I need to take into account the Informix Read-Ahead > > configuration. > > > Any suggestions welcome. Please send responses to email. > > > > > > Thanks, > > > Marc > > > marc.hopkins@aei.com > > > > ********************************************************************** > This email and any files transmitted with it are confidential and > intended solely for the use of the individual or entity to whom they > are addressed. If you have received this email in error please notify > the system manager. > > This footnote also confirms that this email message has been swept by > MIMEsweeper for the presence of computer viruses. > > www.mimesweeper.com > **********************************************************************