Re: Veritas volume manager & OnLine

On Thu, 27 Jul 1995, Jon C. Vemo wrote: } At 02:56 AM 7/27/95 -0400, Edwin Babadaglian wrote: } >In <3v6pa8$30d@usenet.interramp.com> Michael Loftus <mloftus@interramp.com> } writes: } > } >>Greetings! } > } >>I am hoping that someone out there has some experiences to share } >>regarding the use of the Veritas volume manager on a Sparc } >>storage array with OnLine 7.1. } > } >>The use of the volume manager has several big advantages from an } >>administrative standpoint. I'm in the process of getting things } >>set up and am about to begin some experiments of my own. } > } >>I'm wondering if Veritas obscures information from OnLine } >>regarding the physical devices the data is placed on to the point } >>where the effectiveness of PDQ is greatly reduced. In other words, } >>will OnLine realize that a number of logical volumes (mainly simple, } >>non-striped ones) are actually from separate physical devices so } >>that the appropriate number of threads are created for parallel } >>queries? Is it better from a performance standpoint to let OnLine } >>manage the disks and forget about using a volume manager? } > } >>Any and all wisdom greatly appreciated. } > } >>Thanks, } > } >>Mike } > } >>Michael Loftus mloftus@interramp.com } > } >I am not familiar with Veritas' product but I hope it won't matter. } > } >I recall a discussion in the last NYC user group meeting about this } >topic and the general notion is greater awareness can, on the average, } >yield better performance. The usage of a logical volume manager may } >hide the very details that the engine should have for making good } >decisions. But, I think you are confusing threads and i/o requests. The } >idea of a PDQ query is to distribute a single query across processors, } >not disks. On the other hand, the use of a lvm may clash with table } } Yes, you want to spread accross available processors, but the true } performance benefit, comes from being able to access all disks the table } occupies, in parallel. The intent of PDQ (Parallel Data Query), is to } parallelize the query. To do this, use all your processors, and as many } disks as practicle. } } Jon } } >fragmentation since they operate on different levels of abstraction. } > } >I hope this helped. } > } > } >Edwin Babadaglian } >edwinb@panix.com } > } > } ================================================================== } Jon Vemo Internet: jvemo@cyberspace.com } Bothell, WA USA } ------------------------------------------------------------------ } Watch out for road-kill on the information superhighway....SPLAT!! } ================================================================== } } I agree with your "intent of PDQ", but PDQ only addresses the processors. You need to employ some sort of fragmentation, not PDQ, to keep multiple disks busy. With regard to disk performance, it would be fair to consider the LVM as a generalist and the engine as a specialist. Edwin Babadaglian edwinb@panix.com