MySQL server and RAID arrays

[Phaethar]

Hey all,

Looking for a little advice here. We're looking to replace a MySQL system, and are trying to go as fast as possible while still being fault tolerant. So, what we're looking at is a RAID5 array running on U320 15k RPM SCSI drives. I'm curious to know if picking up a dual channel RAID controller will help any, or if a single channel would do just as well.

I'm looking at the LSI Logic MegaRAID cards, 320-1 and 320-2. This will be a Linux system, and I've used LSI Logic cards before, so it's come down to those 2 cards.

I've heard that running RAID5 on a heavily used database system will hinder performance some, so I'm trying to find the best combination of fault tolerance and performance. Would I be better off running the OS on a RAID1 array, and then setting up the MySQL databases on a seperate, RAID5 array? Would running everything on a RAID10 or RAID50 array be better for this kind of application? This sytem will be used only for MySQL, and must be always up as the business depends on it.

Any advice would really be appreciated on this.

Thanks.

 

[DougLite]

How much capacity do you need, what's your budget, and how will you be backing it up?

RAID 0+1 if you are going to have any sort of writes at all. RAID-5 parity calcs, even hardware assisted, will make writes very slow, and rebuilds when a drive goes out will eat into uptime, neither of which is a problem for RAID 0+1.

Dual channel will allow you to extend fault tolerance past your cable/backplane - if a cable pops loose on one of the stripe sets in a 0+1 array, it will remain online. Dual channel would be a good idea if you can afford it.

I would keep the OS on RAID-1, and 10K drives should be adequate for it - you don't anticipate booting it very often. Have the database on a 0+1 array - your looking at 2 10K drives and 4 15K drives.

 

[unhappy_mage]

How much capacity do you need, what's your budget, and how will you be backing it up?

QFT

Raid5 is very looked down upon in the db server world. Avoid if it's at all possible. 0+1 is the way to go here, no doubt about it. Even if you have to start with half as many disks for now and expand later, it's better to start with the proper type of array.

Backups are worth their weight in gold. And no, raid1 doesn't count as backup.

 

[Phaethar]

How much capacity do you need, what's your budget, and how will you be backing it up?

Capacity requirements right now are probably around 140 to 150 GB, which gives us a lot of room to expand. Budget isn't up to me unfortunately, but if presented properly, could potentially be pretty high.

As for backups, we currently have multiple slave systems running, both on and off site. From those slaves, we also create archives of each day and hold on to them for 3 months at multiple locations.

Of course, the nice thing about going SCSI is that they will be hot-swappable. So, should 1 drive die, in theory we should be able to swap the dead drive on the fly and rebuild without any downtime on either a RAID5 or RAID 0+1 setup.

Thanks a lot for the great advice guys!

 

[unhappy_mage]

I'd get around 6 74gb drives, and 0+1 them for about 205 gb real space. I'm not sure you'll need a dual channel as opposed to a single, but if uptime is a big priority, there are a few raid cards that support spanning arrays across controllers, so you might want to look into getting two controllers. The controllers wouldn't be hotswappable, but it'd probably make reconfiguring the array easier if you still have 1 working controller.

Glad to hear you've got backup implemented.

 

[DougLite]

Capacity requirements right now are probably around 140 to 150 GB, which gives us a lot of room to expand. Budget isn't up to me unfortunately, but if presented properly, could potentially be pretty high.

As for backups, we currently have multiple slave systems running, both on and off site. From those slaves, we also create archives of each day and hold on to them for 3 months at multiple locations.

Of course, the nice thing about going SCSI is that they will be hot-swappable. So, should 1 drive die, in theory we should be able to swap the dead drive on the fly and rebuild without any downtime on either a RAID5 or RAID 0+1 setup.

Thanks a lot for the great advice guys!

4x 147GB 15K drives in 0+1 give you 294GB, unformatted. 4x 73GB is going to be right on your capacity requirements - 146GB unformatted. If you opt for 73GB drives, you can economize by going with the previous generation Atlas 15K, Seagate 15K.3 or Fujitsu MAS, which are substantially cheaper because of their lower density, and perform nearly the same as current 15K drives (Atlas 15K II, Seagate 15K.4, Fujitsu MAU), owing to the higher density and fixed reliability requirements limiting performance gains on the newer 147GB 15K drives. These small performance gains are for the expected multi user loads - there just were not that many gains in server performance with the latest crop of enterprise drives, both in 10K and 15K. Single user perfomance increased by incredible amounts, but that doesn't help you. Current 15K drives that meet your capacity requirements are VERY expensive...Your budget may limit you to the nearly as fast second generation drives.

We all know that in the SCSI arena, Seagate is the defacto standard, but Fujitsu and Maxtor drives are faster, and backed by the same 5 year warranties.

Yes, the SCSI drives are hot swappable BUT a RAID-5 rebuild is still a lengthy process. Many controllers allow the array to be accessible while in a degraded state, but RAID-5 rebuilds eat into uptime, hot swap or not. RAID 0+1 simply copies the data to the repalcement drive from its mirror and has it good to in short order. No parity calcs, etc.

Also, you may want to consider a pair of 300GB NearLine drives in RAID-1, such as 'Cuda 7200.8 or MaxLine III, to augment your backup strategy. If you buy Serially Attached SCSI drives, you can attach either of those native SATA-II drives to the same controller as your boot array and DB array.