On m3.xlarge instance (EBS Optimized – True) :

There are two EBS disks: /mnt/data0, /mnt/data1

Single dd:

dd bs=1M count=1024 if=/dev/zero of=/mnt/data0/test conv=fdatasync

1024+0 records in 1024+0 records out 1073741824 bytes (1.1 GB) copied, 15.9016 s, 67.5 MB/s

Two parallel dd's:

1073741824 bytes (1.1 GB) copied, 29.2529 s, 36.7 MB/s

1073741824 bytes (1.1 GB) copied, 33.8585 s, 31.7 MB/s

Clearly, overall EBS throughput is bottleneck. Is it expected?

If total EBS throughput is limited regardless of number of volumes, what is the point in EBS striping?


Looks like there is single network link between instance and EBS cloud. The throughput is described here: http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/EBSOptimized.html

So the answer is, unless IOPS limit is hit, adding more EBS disks won't improve overall EBS performance of the instance.


For me, the main benefit of having multiple smaller EBS volumes rather than one massive one is cost management.

As an example right now on our DEV server, we have 3 x 100 GB EBS volumes attached with ZFS (configured in raidz1 but that's because I'm paranoid about data loss).

The devs keep telling me that we will need upwards of 1 TB worth of space, but until we need it I don't have to provision it. This keeps our costs as low as possible.

  • I would say single 300Gb EBS would be faster, not to mention bigger, than your current setup. Alas, paranoid does not come free. – Ivan Balashov Aug 11 '14 at 16:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.