Last week a DB server lost one of it’s OS partition drives. I should explain that this particular server is a custom build and consists of two arrays run off different controllers. The data storage is run off a 13 drive RAID 5 array. And the OS is run off a 2 drive RAID 1. For whatever reason whoever built this server decided to use old scsi 160 drives for the OS. So when the drive went bad I didn’t have a replacement readily available. What I eventually decided upon was to just replace the both SCSI drives with SATA disks. I was fortunate in that the motherboard has a couple of SATA connectors built in. Though I lost the hardware RAID 1, all the real work is going to be done off the RAID 5 anyway.
I setup the software RAID during the install process for Ubuntu 12.04 Server. It’s not exactly hard but not exactly intuitive either and so I ended up looking around the web just to make sure I knew what I was doing.
Now for the big change. This server has thirteen 750 GB drives which builds out to 7.5 TB. The database that was being run on here took only a fraction of that; somewhere around 15 GB. It seemed to be a waste to use so many spindles just to run a single mid-level usage MySQL instance. What if, instead I hooked this server up as another node in the SAN. Sure it wouldn’t have the fancy features that machines like NetApp and Equallogic come with, but thats not the point. The point is to make the best use out of the resources at hand. The hot spare DB server that took over when I brought this one down, was handling the load with ease and it has far fewer spindles. What better use could I put this machine to?
The configuration I ended up with for the RAID 5 so far includes 6 logical volumes. Two of those volumes host iSCSI luns which are attached to virtual machines which now run the backup MySQL instances for two primary DB servers. Neither of those virtual machines, nor the custom box show much load and I’m confident that should one of the primary databases go down, their virtual machine backup could easily pickup the slack. The other four logical volumes I have attached to VMware VDR virtual machine instances over NFS. The usage with all of six of these virtual machines running their services is now pushing the load of the custom machine pretty much to capacity. I’m now planning to max out the number of hard drives this machine can house in order to create another RAID5. That will allow me to move some of the services over to a different set of spindles.