Disk mirroring offers same performance as Simple volume

Goldmember

This is one of my questions for review on 70-290

I thought Disk mirroring would be slower because information has to be written identically to both disks, whereas a simple volume would be on only one disk.

I guess writing to one disk has the same performance as writing to two disks if the information is the same.

dynamik

Each drive has it's own spindle that will write the data, so they will write simultaneously. It wouldn't take any longer to write to one drive than it would to multiple drives. For example, imagine a few people who can run 100 meters in 15 seconds. Whether it's one person, or all of them side-by-side, the run will still only take 15 seconds. There's no reason to wait for one to finish before the next can go.

Also, remember that RAID-5 is slower because it has to calculate parity information.

Claymoore

It would seem to be slower because you are issuing two write commands instead of one, but it really is only one command to write to both disks. The storage interface stack works a lot like the TCP/IP stack in that there are multiple layers and the different layers don't care what happens after they pass the data on. When you save a file, the application issues a save command to the file system. The file system then changes converts the command into logical addressing that the volume manager can understand and passes the command down to the volume manager. The volume manager converts the command into a physical address to pass on to the disk controller which moves the write heads to the correct platter/track/sector and writes the data. The volume manager (either a hardware chip or OS software) is the only layer that knows the disks are mirrored. After it converts the write command from a logical to physical address, the volume manager just duplicates the command and sends the commands to the disk controller (or controllers in a duplexed system).

Since the disks in a mirrored array are identically laid out in their physical address space, the write heads for each drive would have to be in the same physical location on the disk when they write. Therefore both heads experience the same seek time and rotational latency (the amount of time it takes to move the head across the platter to the correct track and the time it takes for the correct sector to spin around to the head) so both perform the same. On reads however, each head can act independently so you can decrease the rotational latency and seek time based on the position of each head when the read command is issued. The Command Queuing ability of SCSI and fiber channel drives can further increase performance by issuing the read or write commands in physical disc location order rather than the order they were received from the OS in order to reduce the seek time and rotational latency.

I've been studying for the EMC Proven Associate certification - I've learned a lot about disk drives in the last few weeks.