Windows AZURE I/O Tests Temp Drive D: vs Attached Drives

Question

I am testing the speed of the drives on the Windows Azure VM's Drives. It seems to me like the speed difference between the temporary drive d and the attached drives is huge!

The test I have is running iometer program on

Maximum Disk Size 20 Gigs 16 Outstanding I/O's 4k100% read, 0% random

60seconds Run time

Results: Temporary Drive D: Total I/Os per second 60978.94 Drive E (1 30 gig drive): 910.51 Drive F (4 30 gig drives striped together): 899.6

Is this normal?

The reason I am really noticing the difference is in SQL. I basically tried to migrate from my old physical server with sql2000 2 gigs of ram and scsi drives, and that thing is faster than windows azure large image. Faster in that I can run queries about twice as fast.

I turned off diskcaching on os drive in os.

Can someone explain to me what is going on? am I comparing apples to oranges? thanks!

Answer 1

Yes this is normal. The temp disk is a physical disk on the node (only disk I/O here), and the E/F/... disks are persisted disks. This means they are actually page blobs in blob storage and you'll need to take into account network I/O as well.

To improve I/O and throughput you might consider to disable the cache for those disks (this incurs more transaction costs). Read more about this on the Windows Azure Storage Blog: Exploring Windows Azure Drives, Disks, and Images