Chunk writes

Hi, I am in the process of creating the layout for a database. I will have 6 physical disks to work with (plus another 6 for mirroring). I understand that CLEANERS should be set to the number of disks, in this case 6. I also gather from messages that this scheme would perform optimally if I had one chunk on each disk and would have roughly the same number of updates on each. As I will have to work with 9GB disks I will have to place at least 5 chunks per disk. How can I make sure that 5 of the cleaners will not be working on the same disk at one time? How are chunks allocated to the cleaners, by chunk numbers? Which brings me to another question... Does the server look at the pathname of the chunks in it's attempt at trying to sort writes? What I'm trying to get at is whether it makes any difference how you allocate chunks from a performance point of view: is it better to define multiple chunks in one (max 2GB) file with offsets - in which case all of these chunks will share the same pathname - or better to have a separate file (logical volume on AIX) for each chunk? TIA Gabor Heppes <gaborh@au1.ibm.com> -----------== Posted via Deja News, The Discussion Network ==---------- http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own

gheppes@my-dejanews.com wrote: > > Hi, > > I am in the process of creating the layout for a database. I will have 6 > physical disks to work with (plus another 6 for mirroring). I understand that > CLEANERS should be set to the number of disks, in this case 6. I also gather > from messages that this scheme would perform optimally if I had one chunk on > each disk and would have roughly the same number of updates on each. As I > will have to work with 9GB disks I will have to place at least 5 chunks per > disk. How can I make sure that 5 of the cleaners will not be working on the > same disk at one time? How are chunks allocated to the cleaners, by chunk > numbers? > > Which brings me to another question... Does the server look at the pathname > of the chunks in it's attempt at trying to sort writes? What I'm trying to > get at is whether it makes any difference how you allocate chunks from a > performance point of view: is it better to define multiple chunks in one (max > 2GB) file with offsets - in which case all of these chunks will share the > same pathname - or better to have a separate file (logical volume on AIX) for > each chunk? I'm going to again suggest my favorite disk farm configuration, RAID10 (or RAID 1+0). See my post yesterday to "Re: Strping or fragmenting" for details but basically build a large (54GB) stripe from those 6 mirrored pairs (or better reserve one pair for rootdbs/logdbs/tempdbs and use only 5 for the stripe) and partition that into 27 2GB chunks. All the other issues of balancing go away as the stripe firmware/software takes over. Art S. Kagel