RAID (Redundant Array of Independant/Inexpensive Disks) arrays are common in servers and high end workstations. According to Wikipedia, there are 6 official RAID levels, along with several manufacturer specific levels. RAID is available in both hardware and software offerings, with the former being the preferred. Hardware RAID utilises onboard or dedicated RAID controllers, which take the load away from the CPU. They are generally more reliable and better performing than Software RAID - which, as the name suggests, uses Software to create and manage RAID arrays.
As all of my physical and virtual servers here use Microsoft Server Operating Systems (specifically 2003), I decided to research RAID and what was avaiable to me, as my storage is in need of some tweaking. My 2 storage servers, Fileserver and Backupserver both have 4 x 250GB Samsung Spinpoint IDE drives, connected to the mainboard via cheap PCI RAID cards, which only support a handful of RAID levels. As they're not big brand offerings, I am reluctant to use them for creating RAID arrays, in case something goes horribly wrong and I lose GBs of data.
So, I turned to Software RAID and what Server 2003 offers. Microsoft's Technet site has a create article detailing all of the options available to Server Administrators:
- Simple - disks that can be extended, but are usually used as standalone drives
- Spanned - a collection of disks that are combined, to make one larger volume
- Striped - commonly known as RAID0, improves I/O by spreading the volume across 2 disks
- Mirrored - commonly known as RAID1 uses 2 disks to mirror data across them, so if 1 disk dies the data is still available on the 2nd
- RAID5 - uses 3 discs to create a redundant array, so that if 1 disk dies, the array can be rebuilt from the remaining disks
Before I started formatting my ~1TB Backupserver drives, I decided to try out the various options to see which would suit my needs best, which is to combine the 4 x 250GB drives into a larger volume. As the Backupserver keeps a weekly copy of the Fileserver data, redundancy isn't a priority, but as 2 of the 4 drives are nearing their capacity (~232GB formatted) I could do with a larger volume.
Using Microsoft's Virtual Server platform, I was able to create 8 Dynamically Expanding Disks, each of 25GB in size (a 10th of the size of the physical drives). Attaching 2 Virtual SCSI adapters to the Virtual Machine allows you to connect up to 14 drives (7 per adapter), as you can see above. I then switched the Virtual Machine back on and opened up Computer Management:
If you initialise the Disks as Dynamic ones, you can then right click on them and choose which of the available Volume options you want to go with. Above, you can see the options available on Server 2003.
I played around with the Virtual Disks, creating a Striped, Mirrored and RAID-5 array to begin with. The Striped Volume spreads the data over 2 Disks, giving you 2 x 50GB capacity. This effectively improves I/O performance, as you have 2 Disks to read/write to. The Mirrored Volume writes the data to both Disks, giving you just 1 x 25GB capacity, but if one of the Disks fails, you still have a copy of the data. The RAID-5 volume spreads the data over all of the Disks in a way that means if 1 Disk fails, the array can be rebuilt from the remaining Disks. However, if 2 Disks fail, all of the data in the array is lost. In my example, 3 x 25GB drives would give a total capacity of 50GB - one of the drives used for redundancy.
RAID isn't to be used as a backup solution, as other factors can lead to data loss - but some levels provide redundancy in the event of Disk failure and optimise uptime of the system.
To stick with the ~1TB capacity of the 4 x 250GB drives, I opted for the Spanned volume:
In my test environment, I was able to initially create the volume from 4 x 25GB drives, giving me an approximate total of 100GB - which in the Physical servers, would mean near enough 1TB unformatted within a single volume. The beauty of a Spanned volume is that it can easily be extended. So if I find that next year I'm nearing the 1TB capacity, I can chuck another 250GB (or any size) Disk into the Server and extend the Spanned volume onto the new Disk, which would result in a 1.25TB volume. The downside to this is that if 1 of the Disks in the array fails, the whole array is broken and all of the data is lost. However, as I've mentioned, the Backupserver stores a copy of the Fileserver's data, so to have complete data loss, one of the drives in both arrays would need to fail at the same time. Fingers crossed that's something I won't be experiencing.
The performance overhead of Software RAID should be non-existant, as the Spanned volume is simply extending the capacity. Software RAID5 and Mirrored options do put additional strain on the hardware though, which should be taken into consideration if you're thinking of opting for this too.
The next step is to mimic this on the Backupserver, which will require a complete format and converting of all the Disks to Dynamic, then creating the volume and copying the data from the Fileserver across. If all goes to plan, I can then do the same on the Fileserver - which will give me a single 1TB volume, allowing me to re-create the shares without worrying about running out of space anytime soon. To give you an idea of how useful it'll be:
- Documents: 232GB total, 101GB free
- Downloads: 232GB total, 144GB free
- Music: 232GB total, 130GB free
- Video: 232GB total, 69.6GB free