Click here to monitor SSC
SQLServerCentral is supported by Redgate
Log in  ::  Register  ::  Not logged in
Home       Members    Calendar    Who's On

Add to briefcase

Migrating to Amazon - IOPS Question Expand / Collapse
Posted Thursday, December 19, 2013 1:59 PM

Old Hand

Old HandOld HandOld HandOld HandOld HandOld HandOld HandOld Hand

Group: General Forum Members
Last Login: Thursday, September 1, 2016 4:17 PM
Points: 313, Visits: 725
Hi All,

I am working on migrating databases on multiple servers to Amazon. I have a question regarding IOPS. I can determine IOPS based on drives on my SQL Servers which I have done. So for example, the log drive is 1000 IOPS, the Data Drive is 2000 IOPS, and the temp drive is 1000 IOPS. Amazon requests an IOPS setting starting at 1000. I need to determine the amount required. Would it simply be 4000 IOPS or would it be 2000 IOPS? I'm guessing there are multiple trains of thought and I would appreciate any assistance.

I'm also hoping I did the calculations right ((DRIVE IO which is around 125 estimated for 10k drives) * # of drives) * Percent read) +((DRIVE IO which is around 125 estimated for 10k drives) * # of Drives) * IO Penalty)

So say we would have a RAID-5 Array with five 10k Drives and R/W of 50%. That would calculate as ((125*5)*.50) + (((125*5)*.50)*4)

Post #1524792
Posted Tuesday, April 29, 2014 8:30 AM


Group: General Forum Members
Last Login: 2 days ago @ 3:06 AM
Points: 476, Visits: 1,069
I appreciate this is an old post but thought it may be worth replying.

IOPS provisioning is optional and if you provision a set value you will be charged for it. You would be better getting a benchmark from your existing server by capturing the reads \ writes per second across your disks and then setting a value accordingly to your server in AWS.

Are you talking an EC2 AMI or RDS?

In my experience we have found that by not provisioning your IOPS you are at the mercy of amazon and how the underlying disk system services the requests.

For example we have an MySQL RDS instance which on an instance with 1000 provisioned IOPS consistently reports disk write latency at 0.003 seconds when performing approx 700 IOPS.

On another RDS instance running the same workload and an exact copy of the other instance it chokes at a peak of 300 IOPS and disk latency increases by over 600%.

Post #1565960
« Prev Topic | Next Topic »

Add to briefcase

Permissions Expand / Collapse