Thread: 3WARE Card performance boost?
Hi,
I am currently doing large weekly updates with fsync=off. My updates involves SELECT, UPDATE, DELETE and etc. Setting fsync=off works for me since I take a complete backup before the weekly update and run a "sync" and "CHECKPOINT" after each weekly update has completed to ensure the data is all written to disk.
Obviously, I have done this to improve write performance for the update each week. My question is if I install a 3ware or similar card to replace my current software RAID 1 configuration, am I going to see a very large improvement? If so, what would be a ball park figure?
Benjamin Arai
> Obviously, I have done this to improve write performance for the update > each week. My question is if I install a 3ware or similar card to > replace my current software RAID 1 configuration, am I going to see a > very large improvement? If so, what would be a ball park figure? Well that entirely depends on what level... 1. I would suggest LSI 150-6 not 3ware Why? Because 3ware does not make a midrange card that has a battery backed cache :). That is the only reason. 3ware makes good stuff. So anyway... LSI150-6 with Battery Backed cache option. Put 6 drives on it with a RAID 10 array, turn on write cache and you should have a hauling drive. Joshua D. Drake > > *Benjamin Arai* > barai@cs.ucr.edu <mailto:barai@cs.ucr.edu> > http://www.benjaminarai.com <http://www.benjaminarai.com/> > -- The PostgreSQL Company - Command Prompt, Inc. 1.503.667.4564 PostgreSQL Replication, Consulting, Custom Development, 24x7 support Managed Services, Shared and Dedicated Hosting Co-Authors: plPHP, plPerlNG - http://www.commandprompt.com/
My original plan was to buy a 3WARE card and put a 1GB of memory on it to improve writes but I am not sure if that is actually going to help the issue if fsync=off.
From: pgsql-performance-owner@postgresql.org [mailto:pgsql-performance-owner@postgresql.org] On Behalf Of Benjamin Arai
Sent: Wednesday, January 18, 2006 10:10 AM
To: pgsql-performance@postgresql.org
Subject: [PERFORM] 3WARE Card performance boost?Hi,I am currently doing large weekly updates with fsync=off. My updates involves SELECT, UPDATE, DELETE and etc. Setting fsync=off works for me since I take a complete backup before the weekly update and run a "sync" and "CHECKPOINT" after each weekly update has completed to ensure the data is all written to disk.Obviously, I have done this to improve write performance for the update each week. My question is if I install a 3ware or similar card to replace my current software RAID 1 configuration, am I going to see a very large improvement? If so, what would be a ball park figure?Benjamin Arai
http://www.3ware.com/products/serial_ata2-9000.asp Check their data sheet - the cards are BBU ready - all you have to do is order a BBU which you can from here: http://www.newegg.com/Product/Product.asp?Item=N82E16815999601 Alex. On 1/18/06, Joshua D. Drake <jd@commandprompt.com> wrote: > > > Obviously, I have done this to improve write performance for the update > > each week. My question is if I install a 3ware or similar card to > > replace my current software RAID 1 configuration, am I going to see a > > very large improvement? If so, what would be a ball park figure? > > Well that entirely depends on what level... > > 1. I would suggest LSI 150-6 not 3ware > > Why? > > Because 3ware does not make a midrange card that has a battery backed > cache :). That is the only reason. 3ware makes good stuff. > > So anyway... LSI150-6 with Battery Backed cache option. Put 6 drives > on it with a RAID 10 array, turn on write cache and you should have > a hauling drive. > > Joshua D. Drake > > > > > > > *Benjamin Arai* > > barai@cs.ucr.edu <mailto:barai@cs.ucr.edu> > > http://www.benjaminarai.com <http://www.benjaminarai.com/> > > > > > -- > The PostgreSQL Company - Command Prompt, Inc. 1.503.667.4564 > PostgreSQL Replication, Consulting, Custom Development, 24x7 support > Managed Services, Shared and Dedicated Hosting > Co-Authors: plPHP, plPerlNG - http://www.commandprompt.com/ > > ---------------------------(end of broadcast)--------------------------- > TIP 6: explain analyze is your friend >
A 3ware card will re-order your writes to put them more in disk order, which will probably improve performance a bit, but just going from a software RAID 1 to a hardware RAID 1, I would not imagine that you will see much of a performance boost. Really to get better performance you will need to add more drives, or faster drives. If you are currently running 7200 RPM consumer drives, going to a 10000RPM WD Raptor drive will probably increase performance by about 30%, again not all that much. Alex On 1/18/06, Benjamin Arai <barai@cs.ucr.edu> wrote: > > Hi, > > I am currently doing large weekly updates with fsync=off. My updates > involves SELECT, UPDATE, DELETE and etc. Setting fsync=off works for me > since I take a complete backup before the weekly update and run a "sync" and > "CHECKPOINT" after each weekly update has completed to ensure the data is > all written to disk. > > Obviously, I have done this to improve write performance for the update each > week. My question is if I install a 3ware or similar card to replace my > current software RAID 1 configuration, am I going to see a very large > improvement? If so, what would be a ball park figure? > > > Benjamin Arai > barai@cs.ucr.edu > http://www.benjaminarai.com >
On Wed, 2006-01-18 at 10:26 -0800, Benjamin Arai wrote: > My original plan was to buy a 3WARE card and put a 1GB of memory on it > to improve writes but I am not sure if that is actually going to help > the issue if fsync=off. My experience with a 3Ware 9500S-8 card are rather disappointing, especially the write performance of the card, which is extremely poor. Reading is OK. -- Groeten, Joost Kraaijeveld Askesis B.V. Molukkenstraat 14 6524NB Nijmegen tel: 024-3888063 / 06-51855277 fax: 024-3608416 e-mail: J.Kraaijeveld@Askesis.nl web: www.askesis.nl
Benjamin Arai wrote: > Obviously, I have done this to improve write performance for the update > each week. My question is if I install a 3ware or similar card to > replace my current software RAID 1 configuration, am I going to see a > very large improvement? If so, what would be a ball park figure? The key is getting a card with the ability to upgrade the onboard ram. Our previous setup was a LSI MegaRAID 320-1 (128MB), 4xRAID10, fsync=off. Replaced it with a ARC-1170 (1GB) w/ 24x7200RPM SATA2 drives (split into 3 8-drive RAID6 arrays) and performance for us is through the ceiling. For OLTP type updates, we've gotten about +80% increase. For massive 1-statement updates, performance increase is in the +triple digits.
On Wed, Jan 18, 2006 at 01:58:09PM -0800, William Yu wrote: > The key is getting a card with the ability to upgrade the onboard ram. > > Our previous setup was a LSI MegaRAID 320-1 (128MB), 4xRAID10, > fsync=off. Replaced it with a ARC-1170 (1GB) w/ 24x7200RPM SATA2 drives > (split into 3 8-drive RAID6 arrays) and performance for us is through > the ceiling. Well, the fact that you went from four to 24 disks would perhaps be a bigger factor than the amount of RAM... /* Steinar */ -- Homepage: http://www.sesse.net/
Steinar H. Gunderson wrote: > On Wed, Jan 18, 2006 at 01:58:09PM -0800, William Yu wrote: >> The key is getting a card with the ability to upgrade the onboard ram. >> >> Our previous setup was a LSI MegaRAID 320-1 (128MB), 4xRAID10, >> fsync=off. Replaced it with a ARC-1170 (1GB) w/ 24x7200RPM SATA2 drives >> (split into 3 8-drive RAID6 arrays) and performance for us is through >> the ceiling. > > Well, the fact that you went from four to 24 disks would perhaps be a bigger > factor than the amount of RAM... > > /* Steinar */ Actually no. Our 2xOpteron 244 server is NOT fast enough to drive an array this large. That's why we had to split it up into 3 different arrays. I tried all different RAID configs and once past about 8 drives, I got the same performance no matter what because the CPU was pegged at 100%. Right now, 2 of the arrays are just mirroring each other because we can't seem utilize the performance right now. (Also protects against cabling/power supply issues as we're using 3 seperate external enclosures.) The 1GB RAM is much bigger because it almost completely hides the write activity. Looking at iostat while all our jobs are running, there's almost no disk activity. If I manually type "sync", I see 1 big 250MB-500MB write storm for 2 seconds but otherwise, writes just slowly dribble out to disk.
On Wed, Jan 18, 2006 at 10:24:17AM -0800, Joshua D. Drake wrote: > Because 3ware does not make a midrange card that has a battery backed > cache :). That is the only reason. 3ware makes good stuff. Why worry about battery-backup if he's running with fsync off? -- Jim C. Nasby, Sr. Engineering Consultant jnasby@pervasive.com Pervasive Software http://pervasive.com work: 512-231-6117 vcard: http://jim.nasby.net/pervasive.vcf cell: 512-569-9461
On Jan 18, 2006, at 1:09 PM, Benjamin Arai wrote: > Obviously, I have done this to improve write performance for the > update each > week. My question is if I install a 3ware or similar card to > replace my I'll bet that if you increase your checkpoint_segments (and corresponding timeout value) to something in the 10's or higher it will help you a lot.
He's talking about RAID 1 here, not a gargantuan RAID 6. Onboard RAM on the controller card is going to make very little difference. All it will do is allow the card to re-order writes to a point (not all cards even do this). Alex. On 1/18/06, William Yu <wyu@talisys.com> wrote: > Benjamin Arai wrote: > > Obviously, I have done this to improve write performance for the update > > each week. My question is if I install a 3ware or similar card to > > replace my current software RAID 1 configuration, am I going to see a > > very large improvement? If so, what would be a ball park figure? > > The key is getting a card with the ability to upgrade the onboard ram. > > Our previous setup was a LSI MegaRAID 320-1 (128MB), 4xRAID10, > fsync=off. Replaced it with a ARC-1170 (1GB) w/ 24x7200RPM SATA2 drives > (split into 3 8-drive RAID6 arrays) and performance for us is through > the ceiling. > > For OLTP type updates, we've gotten about +80% increase. For massive > 1-statement updates, performance increase is in the +triple digits. > > ---------------------------(end of broadcast)--------------------------- > TIP 2: Don't 'kill -9' the postmaster >