Read-ahead parameters
IDS 7.30.uc7 HPUX 10.20 The bugs are supposed to be gone; after eight months I'm ready to reenable read-aheads. Our original values of RA_PAGES / RA_THRESHOLD were 150 / 75. These seemed a bit high based on the postings that I've seen since that time. We're an OLTP shop, with batch (report) processing thrown in for good measure. All of our queries run on some index; we have no large sequential table scans. Our disks are 4GB FW SCSI. I'm planning to start at values of 64 / 32. Any thoughts, comments, or experiences? John Carlson Informix DBA WHSmith USA
Carlson@WHSmith wrote: > > IDS 7.30.uc7 > HPUX 10.20 > > The bugs are supposed to be gone; after eight months I'm ready to > reenable read-aheads. > > Our original values of RA_PAGES / RA_THRESHOLD were 150 / 75. These > seemed a bit high based on the postings that I've seen since that time. > We're an OLTP shop, with batch (report) processing thrown in for good > measure. All of our queries run on some index; we have no large > sequential table scans. Our disks are 4GB FW SCSI. > > I'm planning to start at values of 64 / 32. Any thoughts, comments, or > experiences? It depends. If your disk farm is slowish: singleton disks with multiple Informix chunks per drive and spindle speed <7200RPM then those values will be about right. Blazing disk farm made of 20 pair RAID10 sets then you may well get away very well with 16/4. Art S. Kagel
20-pair RAID 10? Not yet . . . . maybe one day . . . /* dream sequence */ Wow! No response problems on your system, huh? Thanks for the response. John Carlson Informix DBA WHSmith USA Art S. Kagel wrote: > > Carlson@WHSmith wrote: > > > > IDS 7.30.uc7 > > HPUX 10.20 > > > > The bugs are supposed to be gone; after eight months I'm ready to > > reenable read-aheads. > > > > Our original values of RA_PAGES / RA_THRESHOLD were 150 / 75. These > > seemed a bit high based on the postings that I've seen since that time. > > We're an OLTP shop, with batch (report) processing thrown in for good > > measure. All of our queries run on some index; we have no large > > sequential table scans. Our disks are 4GB FW SCSI. > > > > I'm planning to start at values of 64 / 32. Any thoughts, comments, or > > experiences? > > It depends. If your disk farm is slowish: singleton disks with > multiple Informix chunks per drive and spindle speed <7200RPM then > those values will be about right. Blazing disk farm made of 20 pair > RAID10 sets then you may well get away very well with 16/4. > > Art S. Kagel
Carlson@WHSmith wrote: > > 20-pair RAID 10? Not yet . . . . maybe one day . . . /* dream sequence > */ > Wow! No response problems on your system, huh? > > Thanks for the response. Can't have any! The Bloomberg IPC environment enforces a 10 second response time to all interactive user requests or the user gets a "timeout" message which is usually followed by our help-desk getting a call. So yes, response problems are rare, migrane headaches are not when things get close. > John Carlson > Informix DBA > WHSmith USA > > Art S. Kagel wrote: > > > > Carlson@WHSmith wrote: > > > > > > IDS 7.30.uc7 > > > HPUX 10.20 > > > > > > The bugs are supposed to be gone; after eight months I'm ready to > > > reenable read-aheads. > > > > > > Our original values of RA_PAGES / RA_THRESHOLD were 150 / 75. These > > > seemed a bit high based on the postings that I've seen since that time. > > > We're an OLTP shop, with batch (report) processing thrown in for good > > > measure. All of our queries run on some index; we have no large > > > sequential table scans. Our disks are 4GB FW SCSI. > > > > > > I'm planning to start at values of 64 / 32. Any thoughts, comments, or > > > experiences? > > > > It depends. If your disk farm is slowish: singleton disks with > > multiple Informix chunks per drive and spindle speed <7200RPM then > > those values will be about right. Blazing disk farm made of 20 pair > > RAID10 sets then you may well get away very well with 16/4. > > > > Art S. Kagel
So I'd have to ask - do your requests cause writes into the database, and if so how do you get less than 10 second response time with checkpoints blocking all writers? "Art S. Kagel" wrote: > Carlson@WHSmith wrote: > > > > 20-pair RAID 10? Not yet . . . . maybe one day . . . /* dream sequence > > */ > > Wow! No response problems on your system, huh? > > > > Thanks for the response. > > Can't have any! The Bloomberg IPC environment enforces a 10 second > response time to all interactive user requests or the user gets a > "timeout" message which is usually followed by our help-desk getting a > call. So yes, response problems are rare, migrane headaches are not > when things get close. > > > John Carlson > > Informix DBA > > WHSmith USA > > > > Art S. Kagel wrote: > > > > > > Carlson@WHSmith wrote: > > > > > > > > IDS 7.30.uc7 > > > > HPUX 10.20 > > > > > > > > The bugs are supposed to be gone; after eight months I'm ready to > > > > reenable read-aheads. > > > > > > > > Our original values of RA_PAGES / RA_THRESHOLD were 150 / 75. These > > > > seemed a bit high based on the postings that I've seen since that time. > > > > We're an OLTP shop, with batch (report) processing thrown in for good > > > > measure. All of our queries run on some index; we have no large > > > > sequential table scans. Our disks are 4GB FW SCSI. > > > > > > > > I'm planning to start at values of 64 / 32. Any thoughts, comments, or > > > > experiences? > > > > > > It depends. If your disk farm is slowish: singleton disks with > > > multiple Informix chunks per drive and spindle speed <7200RPM then > > > those values will be about right. Blazing disk farm made of 20 pair > > > RAID10 sets then you may well get away very well with 16/4. > > > > > > Art S. Kagel
I have read somewhere that the max value for read ahead is 32. I do not remember where and I cannot find a reference to a max value in any doc. Any references? Doug McAllister Fidelity Investments "Carlson@WHSmith" wrote: > IDS 7.30.uc7 > HPUX 10.20 > > The bugs are supposed to be gone; after eight months I'm ready to > reenable read-aheads. > > Our original values of RA_PAGES / RA_THRESHOLD were 150 / 75. These > seemed a bit high based on the postings that I've seen since that time. > We're an OLTP shop, with batch (report) processing thrown in for good > measure. All of our queries run on some index; we have no large > sequential table scans. Our disks are 4GB FW SCSI. > > I'm planning to start at values of 64 / 32. Any thoughts, comments, or > experiences? > > John Carlson > Informix DBA > WHSmith USA