Thread: advice on raid controller
Hi, i'm on the verge of buying a "MegaRAID SCSI 320-2" raid controller. I need it to build a db server using 4x ultra320 scsi disks i'm thinking raid 1+0 but will try with raid5 too and compare Does anyone have any experience with this model, good or bad i'd like to know.. thanks :) as seen: http://uk.azzurri.com/product/product.cgi?productId=188 Regards, Richard. PS: whoever mentioned starting a site with raid controller reviews, excellent idea - its hard to find decent info on which card to buy.
On 2003-09-27T18:24:33+0100, Richard Jones wrote: > i'm on the verge of buying a "MegaRAID SCSI 320-2" raid controller. You may want to check out the PCI-X version of this controller that LSILogic just released (MegaRAID SCSI 320-2X). PCI-X is backwards compatible with PCI, but also gives you greater bandwidth if your motherboard supports it (at least, that's the marketing fluff). Adaptec and Intel makes (PCI) controllers with similar specs to the one you mentioned. > I need it to build a db server using 4x ultra320 scsi disks > i'm thinking raid 1+0 but will try with raid5 too and compare The Fujitsu 15k drives look sweet :-) > PS: whoever mentioned starting a site with raid controller reviews, excellent > idea - its hard to find decent info on which card to buy. You may want to check recent archives for RAID threads. /Allan -- Allan Wind P.O. Box 2022 Woburn, MA 01888-0022 USA
On Sat, 2003-09-27 at 12:24, Richard Jones wrote: > Hi, i'm on the verge of buying a "MegaRAID SCSI 320-2" raid controller. > I need it to build a db server using 4x ultra320 scsi disks > i'm thinking raid 1+0 but will try with raid5 too and compare > > Does anyone have any experience with this model, good or bad i'd like to > know.. thanks :) > > as seen: > http://uk.azzurri.com/product/product.cgi?productId=188 I don't see anything on that page regarding RAM cache. It's been my experience that RAID 5 needs a *minimum* of 128MB cache to have good performance. -- ----------------------------------------------------------------- Ron Johnson, Jr. ron.l.johnson@cox.net Jefferson, LA USA Why is cyber-crime not being effectively controlled? What is fuelling the rampancy? * Parental apathy & the public education system http://www.linuxsecurity.com/feature_stories/feature_story-150.html
RIchard, > Hi, i'm on the verge of buying a "MegaRAID SCSI 320-2" raid controller. > I need it to build a db server using 4x ultra320 scsi disks > i'm thinking raid 1+0 but will try with raid5 too and compare Depends on your type of database. If you're doing web or OLAP (lots of read-only queries) RAID 5 will probably be better. If you're doing OLTP (lots of read-write) RAID 10 will almost certainly be better. But if you have time, testing is always best. > as seen: > http://uk.azzurri.com/product/product.cgi?productId=188 I haven'te used it personally, but what I don't see in the docs is a battery-backed cache. Without battery backup on the write cache, IMHO you are better off with Linux of BSD software RAID, since you'll have to turn off the card's write cache, lest your database get corrupted on power-out. -- Josh Berkus Aglio Database Solutions San Francisco
As others have mentioned, you really ought to get battery-backed cache if you're doing any volume of writes. The ability to do safe write-back caching makes an *insane* difference to write performance. The site you link to also has that for only 15% more money: http://uk.azzurri.com/product/product.cgi?productId=80 No experience with the card(s) I'm afraid. In general though, U320 will only be faster than U160 for large sequential reads, or when you have silly numbers of disks on a channel (i.e. more than 4/channel). If you have silly numbers of disks, then RAID5 will probably be better, if you have 4 disks total then RAID1+0 will probably be better. In between it depends on all sorts of other factors. Bear in mind though that if you *do* have silly numbers of disks then more channels and more cache will count for more than anything else, so spend the money on that rather than latest-and-greatest performance for a single channel. HTH Matt > -----Original Message----- > From: pgsql-performance-owner@postgresql.org > [mailto:pgsql-performance-owner@postgresql.org]On Behalf Of Richard > Jones > Sent: 27 September 2003 18:25 > To: pgsql-performance@postgresql.org > Subject: [PERFORM] advice on raid controller > > > Hi, i'm on the verge of buying a "MegaRAID SCSI 320-2" raid controller. > I need it to build a db server using 4x ultra320 scsi disks > i'm thinking raid 1+0 but will try with raid5 too and compare > > Does anyone have any experience with this model, good or bad i'd like to > know.. thanks :) > > as seen: > http://uk.azzurri.com/product/product.cgi?productId=188 > > Regards, > Richard. > > PS: whoever mentioned starting a site with raid controller > reviews, excellent > idea - its hard to find decent info on which card to buy. > > > ---------------------------(end of broadcast)--------------------------- > TIP 9: the planner will ignore your desire to choose an index scan if your > joining column's datatypes do not match >
I've used the megaraid / LSI cards in the past and they were pretty good in terms of reliability, but the last one I used was the 328 model, from 4 years ago or so. that one had a battery backup option for the cache, and could go to 128 Meg. We tested it with 4/16 and 128 meg ram, and it was about the same with each, but we didn't do heavy parallel testing either. Here's the page on the megaraid cards at lsilogic.com: http://www.lsilogic.com/products/stor_prod/raid/ultra320products.html On Sun, 28 Sep 2003, Matt Clark wrote: > As others have mentioned, you really ought to get battery-backed cache if > you're doing any volume of writes. The ability to do safe write-back > caching makes an *insane* difference to write performance. > > The site you link to also has that for only 15% more money: > http://uk.azzurri.com/product/product.cgi?productId=80 > > No experience with the card(s) I'm afraid. > > In general though, U320 will only be faster than U160 for large sequential > reads, or when you have silly numbers of disks on a channel (i.e. more than > 4/channel). If you have silly numbers of disks, then RAID5 will probably be > better, if you have 4 disks total then RAID1+0 will probably be better. In > between it depends on all sorts of other factors. Bear in mind though that > if you *do* have silly numbers of disks then more channels and more cache > will count for more than anything else, so spend the money on that rather > than latest-and-greatest performance for a single channel. > > HTH > > Matt > > > -----Original Message----- > > From: pgsql-performance-owner@postgresql.org > > [mailto:pgsql-performance-owner@postgresql.org]On Behalf Of Richard > > Jones > > Sent: 27 September 2003 18:25 > > To: pgsql-performance@postgresql.org > > Subject: [PERFORM] advice on raid controller > > > > > > Hi, i'm on the verge of buying a "MegaRAID SCSI 320-2" raid controller. > > I need it to build a db server using 4x ultra320 scsi disks > > i'm thinking raid 1+0 but will try with raid5 too and compare > > > > Does anyone have any experience with this model, good or bad i'd like to > > know.. thanks :) > > > > as seen: > > http://uk.azzurri.com/product/product.cgi?productId=188 > > > > Regards, > > Richard. > > > > PS: whoever mentioned starting a site with raid controller > > reviews, excellent > > idea - its hard to find decent info on which card to buy. > > > > > > ---------------------------(end of broadcast)--------------------------- > > TIP 9: the planner will ignore your desire to choose an index scan if your > > joining column's datatypes do not match > > > > > ---------------------------(end of broadcast)--------------------------- > TIP 2: you can get off all lists at once with the unregister command > (send "unregister YourEmailAddressHere" to majordomo@postgresql.org) >
On Mon, 2003-09-29 at 06:48, scott.marlowe wrote: > I've used the megaraid / LSI cards in the past and they were pretty good > in terms of reliability, but the last one I used was the 328 model, from 4 > years ago or so. that one had a battery backup option for the cache, and > could go to 128 Meg. We tested it with 4/16 and 128 meg ram, and it was > about the same with each, but we didn't do heavy parallel testing either. > > Here's the page on the megaraid cards at lsilogic.com: > > http://www.lsilogic.com/products/stor_prod/raid/ultra320products.html > > On Sun, 28 Sep 2003, Matt Clark wrote: > > > As others have mentioned, you really ought to get battery-backed cache if > > you're doing any volume of writes. The ability to do safe write-back > > caching makes an *insane* difference to write performance. > > > > The site you link to also has that for only 15% more money: > > http://uk.azzurri.com/product/product.cgi?productId=80 > > > > No experience with the card(s) I'm afraid. > > > > In general though, U320 will only be faster than U160 for large sequential > > reads, or when you have silly numbers of disks on a channel (i.e. more than > > 4/channel). If you have silly numbers of disks, then RAID5 will probably be > > better, if you have 4 disks total then RAID1+0 will probably be better. In > > between it depends on all sorts of other factors. Bear in mind though that > > if you *do* have silly numbers of disks then more channels and more cache > > will count for more than anything else, so spend the money on that rather > > than latest-and-greatest performance for a single channel. Just to add my thoughts, we use the MegaRaid Elite 1650 in 3 servers here that drive our core databases. We paired up the controllers with the Seagate Cheetah 10k drives, we could have purchased the X15's which are Seagate's 15k version, but due to budget constraints and lack of true performance increase from the 10k to the 15k rpm drives we didn't opt for them. I have to say that I've been 100% pleased with the performance and reliability of the Megaraid controllers. They do everything a good controller should and they are very easy to manage. The driver is actively maintained by the guys at LSI and their tech support personnel are very good as well. If you want any specific information or have any questions about our experience or configuration please feel free to contact me. Sincerely, Will LaShell > > HTH > > > > Matt > > > > > -----Original Message----- > > > From: pgsql-performance-owner@postgresql.org > > > [mailto:pgsql-performance-owner@postgresql.org]On Behalf Of Richard > > > Jones > > > Sent: 27 September 2003 18:25 > > > To: pgsql-performance@postgresql.org > > > Subject: [PERFORM] advice on raid controller > > > > > > > > > Hi, i'm on the verge of buying a "MegaRAID SCSI 320-2" raid controller. > > > I need it to build a db server using 4x ultra320 scsi disks > > > i'm thinking raid 1+0 but will try with raid5 too and compare > > > > > > Does anyone have any experience with this model, good or bad i'd like to > > > know.. thanks :) > > > > > > as seen: > > > http://uk.azzurri.com/product/product.cgi?productId=188 > > > > > > Regards, > > > Richard. > > > > > > PS: whoever mentioned starting a site with raid controller > > > reviews, excellent > > > idea - its hard to find decent info on which card to buy.
Attachment
On 29 Sep 2003, Will LaShell wrote: > On Mon, 2003-09-29 at 06:48, scott.marlowe wrote: > > I've used the megaraid / LSI cards in the past and they were pretty good > > in terms of reliability, but the last one I used was the 328 model, from 4 > > years ago or so. that one had a battery backup option for the cache, and > > could go to 128 Meg. We tested it with 4/16 and 128 meg ram, and it was > > about the same with each, but we didn't do heavy parallel testing either. > > > > Here's the page on the megaraid cards at lsilogic.com: > > > > http://www.lsilogic.com/products/stor_prod/raid/ultra320products.html > > > > On Sun, 28 Sep 2003, Matt Clark wrote: > > > > > As others have mentioned, you really ought to get battery-backed cache if > > > you're doing any volume of writes. The ability to do safe write-back > > > caching makes an *insane* difference to write performance. > > > > > > The site you link to also has that for only 15% more money: > > > http://uk.azzurri.com/product/product.cgi?productId=80 > > > > > > No experience with the card(s) I'm afraid. > > > > > > In general though, U320 will only be faster than U160 for large sequential > > > reads, or when you have silly numbers of disks on a channel (i.e. more than > > > 4/channel). If you have silly numbers of disks, then RAID5 will probably be > > > better, if you have 4 disks total then RAID1+0 will probably be better. In > > > between it depends on all sorts of other factors. Bear in mind though that > > > if you *do* have silly numbers of disks then more channels and more cache > > > will count for more than anything else, so spend the money on that rather > > > than latest-and-greatest performance for a single channel. > > Just to add my thoughts, we use the MegaRaid Elite 1650 in 3 servers > here that drive our core databases. We paired up the controllers with > the Seagate Cheetah 10k drives, we could have purchased the X15's which > are Seagate's 15k version, but due to budget constraints and lack of > true performance increase from the 10k to the 15k rpm drives we didn't > opt for them. > > I have to say that I've been 100% pleased with the performance and > reliability of the Megaraid controllers. They do everything a good > controller should and they are very easy to manage. The driver is > actively maintained by the guys at LSI and their tech support personnel > are very good as well. > > If you want any specific information or have any questions about our > experience or configuration please feel free to contact me. To add one more feature the LSI/MegaRAIDs have that I find interesting, you can put two in a machine, build a RAID0 or 5 on each card, then mirror the two cards together, and should one card / RAID0 ot 5 chain die, the other card will keep working. I.e. the work like one big card with failover.
Stupid question, perhaps, but would a battery-backed cache make it safe to set fsync=false in postgresql.conf? /Palle --On söndag, september 28, 2003 13.07.57 +0100 Matt Clark <matt@ymogen.net> wrote: > As others have mentioned, you really ought to get battery-backed cache if > you're doing any volume of writes. The ability to do safe write-back > caching makes an *insane* difference to write performance.
Come to think of it, I guess a battery-backed cache will make fsync as fast as no fsync, right? So, the q was kinda stoopid... :-/ /Palle --On måndag, september 29, 2003 23.31.54 +0200 Palle Girgensohn <girgen@pingpong.net> wrote: > Stupid question, perhaps, but would a battery-backed cache make it safe > to set fsync=false in postgresql.conf? > > /Palle > > --On söndag, september 28, 2003 13.07.57 +0100 Matt Clark > <matt@ymogen.net> wrote: > >> As others have mentioned, you really ought to get battery-backed cache if >> you're doing any volume of writes. The ability to do safe write-back >> caching makes an *insane* difference to write performance. > > > >
Not in general. Besides, with a write-back cache an fsync() is very nearly 'free', as the controller will report the write as completed as soon as it's written to cache. I keep meaning to benchmark the difference, but I only have the facility on a production box, so caution gets the better of me every time :-) AFAIK the fsync calls are used to guarantee the _ordering_ of writes to permanent storage (i.e. fsync() is called before doing something, rather than after doing something. So PG can be sure that before it does B, A has definitely been written to disk). But I could well be wrong. And there could well be strategies exploitable with the knowledge that a write-back cache exists that aren't currently implemented - though intuitively I doubt it. M > -----Original Message----- > From: Palle Girgensohn [mailto:girgen@pingpong.net] > Sent: 29 September 2003 22:32 > To: Matt Clark; rj@last.fm; pgsql-performance@postgresql.org > Subject: Re: [PERFORM] advice on raid controller > > > Stupid question, perhaps, but would a battery-backed cache make > it safe to > set fsync=false in postgresql.conf? > > /Palle > > --On söndag, september 28, 2003 13.07.57 +0100 Matt Clark > <matt@ymogen.net> > wrote: > > > As others have mentioned, you really ought to get > battery-backed cache if > > you're doing any volume of writes. The ability to do safe write-back > > caching makes an *insane* difference to write performance. > > > > >
\Palle Girgensohn wrote: > Come to think of it, I guess a battery-backed cache will make fsync as fast > as no fsync, right? So, the q was kinda stoopid... :-/ With fsync off, the data might never get to the battery-backed RAM. :-( -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup. | Newtown Square, Pennsylvania 19073
>>>>> "PG" == Palle Girgensohn <girgen@pingpong.net> writes: PG> Come to think of it, I guess a battery-backed cache will make fsync as PG> fast as no fsync, right? So, the q was kinda stoopid... :-/ In my testing, yes, the battery cache makes fsync=true just about as fast as fsync=false. it was only about 2 seconds slower (out of 4 hours) while doing a restore. -- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Vivek Khera, Ph.D. Khera Communications, Inc. Internet: khera@kciLink.com Rockville, MD +1-240-453-8497 AIM: vivekkhera Y!: vivek_khera http://www.khera.org/~vivek/
>>>>> "RJ" == Richard Jones <rj@last.fm> writes: RJ> Hi, i'm on the verge of buying a "MegaRAID SCSI 320-2" raid controller. RJ> I need it to build a db server using 4x ultra320 scsi disks RJ> i'm thinking raid 1+0 but will try with raid5 too and compare No specific tips on that particular RAID, but in general it seems that you want to *disable* the read-ahead and enable the write-back cache. This is from reading on the linux megaraid developers list. Also, for 4 disks, go with RAID 1+0 for your best performance. I found it faster. However, with my 14 disk system, RAID5 is fastest. -- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Vivek Khera, Ph.D. Khera Communications, Inc. Internet: khera@kciLink.com Rockville, MD +1-240-453-8497 AIM: vivekkhera Y!: vivek_khera http://www.khera.org/~vivek/