Thread: Hard Drive Usage for Speeding up Big Queries
We are trying to optimize our Database server without spending a fortune on hardware. Here is our current setup. Main Drive array: 8x 750 GB SATA 2 Drives in a RAID 10 Configuration, this stores the OS, Applications, and PostgreSQL Data drives. 3 TB Array, 2 TB Parition for PostgreSQL. Secondary drive array: 2x 36 GB SAS 15000 RPM Drives in a RAID 1 Configuration: the pg_xlog directory, checkpoints set to use about 18 GB max, this way when massive numbers of small writes occur, they don't slow the system down. Drive failure loses no data. Checkpoints will be another matter, hope to keep under control with bgwriter tweaking. Now our "normal" activities are really fast. Scanning data, etc., all runs pretty quickly. What is NOT fast is some of the massive queries. We have some Updates with joins that match a 100m line table with a 200m line table. Outside of custom coding pl/pgsql code that creates the subfunction on the fly (which is up for consideration) to try to keep the Matching joins to an O(n) problem from the current O(n^2) one, we are looking at hardware as an option to help speed up these big batch queries that sometimes run for 5-6 days. CPU not a problem, 2x Quad-core Xeon, never taxing more than 13%, this will change as more of our database functions are brought over here from the other servers RAM is not upgradable, have 48GB of RAM on there. Work_mem shouldn't be the issue, the big processes get Work_mem set to 10GB, and if they are using temp tables, another 6-8GB for temp_buffers. Maintenance Mem is set to 2 GB. However, the joins of two 50GB tables really just can't be solved in RAM without using drive space. My question is, can hardware speed that up? Would putting a 400 GB SAS Drive (15000 RPM) in just to handle PostgreSQL temp files help? Considering it would store "in process" queries and not "completed transactions" I see no reason to mirror the drive. If it fails, we'd simply unmount it, replace it, then remount it, it could use the SATA space in the mean time. Would that speed things up, and if so, where in the drive mappings should that partition go? Thank you for your help. I'm mostly interested in if I can speed these things up from 5-6 days to < 1 day, otherwise I need to look at optimizing it. Alex
On Jan 28, 2008 7:54 AM, Alex Hochberger <alex@dsgi.us> wrote: > We are trying to optimize our Database server without spending a > fortune on hardware. Here is our current setup. > > Main Drive array: 8x 750 GB SATA 2 Drives in a RAID 10 Configuration, > this stores the OS, Applications, and PostgreSQL Data drives. 3 TB > Array, 2 TB Parition for PostgreSQL. > Secondary drive array: 2x 36 GB SAS 15000 RPM Drives in a RAID 1 > Configuration: the pg_xlog directory, checkpoints set to use about 18 > GB max, this way when massive numbers of small writes occur, they > don't slow the system down. Drive failure loses no data. Checkpoints > will be another matter, hope to keep under control with bgwriter > tweaking. > SNIP > However, the joins of two 50GB tables really just can't be solved in > RAM without using drive space. My question is, can hardware speed > that up? Would putting a 400 GB SAS Drive (15000 RPM) in just to > handle PostgreSQL temp files help? Considering it would store "in > process" queries and not "completed transactions" I see no reason to > mirror the drive. If it fails, we'd simply unmount it, replace it, > then remount it, it could use the SATA space in the mean time. > > Would that speed things up, and if so, where in the drive mappings > should that partition go? Do you have a maintenance window to experiment in? Try putting it on the pg_xlog array to see if it speeds up the selects during one. Then you'll know. I'm thinking it will help a little, but there's only so much you can do with 50g result sets.
On Jan 28, 2008 8:54 AM, Alex Hochberger <alex@dsgi.us> wrote: > We are trying to optimize our Database server without spending a > fortune on hardware. Here is our current setup. > > Main Drive array: 8x 750 GB SATA 2 Drives in a RAID 10 Configuration, > this stores the OS, Applications, and PostgreSQL Data drives. 3 TB > Array, 2 TB Parition for PostgreSQL. > Secondary drive array: 2x 36 GB SAS 15000 RPM Drives in a RAID 1 > Configuration: the pg_xlog directory, checkpoints set to use about 18 > GB max, this way when massive numbers of small writes occur, they > don't slow the system down. Drive failure loses no data. Checkpoints > will be another matter, hope to keep under control with bgwriter > tweaking. > > Now our "normal" activities are really fast. Scanning data, etc., all > runs pretty quickly. What is NOT fast is some of the massive > queries. We have some Updates with joins that match a 100m line table > with a 200m line table. Outside of custom coding pl/pgsql code that > creates the subfunction on the fly (which is up for consideration) to > try to keep the Matching joins to an O(n) problem from the current > O(n^2) one, we are looking at hardware as an option to help speed up > these big batch queries that sometimes run for 5-6 days. > Well, you have already put some thought into your hardware...but the awful truth is that the sata drives are just terrible at seeking once you start seeing significant numbers of page faults to disk, and getting killed on sorting on top of it. Maybe the best plan of attack here is to post some explain times, and the relevant query. Perhaps there are some optimizations in indexing strategies or other query tactics (the pl/pgsql function smells suspicious as you have already noted), and hopefully the giant sort can be optimized out. You have a nasty problem that may require some out of the box thinking, so the more information you can provide the better. merlin