locked
Performance/costs improvements from single SQL database to elastic scale RRS feed

  • Question

  • Hi there!

    I already posted a previous question on whats the best way to setup a elastic scale architecture for our application(https://social.msdn.microsoft.com/Forums/azure/en-US/82fabac7-137b-46d6-a9f0-5e71e4bbc9eb/using-datadependent-routing-in-combination-with-membership-provider?forum=ssdsgetstarted).

    The next thing we need to know before we can implement elastic scale is if this is really going to help us with our performance and costs where we experiencing problems at the moment.

    Currently we have a single SQL database(p3 800 DTU's).

    We have run queries against our database which can calculate the number of DTU's we need to get good performance for that database.

    My questions are:

    1.When we implement elastic scale, can we really improve our performance and can this lead to saving costs on SQL databases?

    2. Is there maybe a way we can easily test this by running queries or setting up an environment  where we can see real differences in DTU's per Shard?

    Thanks!

    Wednesday, February 11, 2015 9:30 AM

Answers

  • Hi Elmar,

    If you're already hitting performance limits with your P3 database, other than upgrading to the new V12 server for improved premium performance, that is the highest you can currently vertically scale.  Thus, it becomes advantageous to scale out to achieve better performance. 

    A small caveat, my answers below are really contingent upon your workload, query patterns, and sharding scheme.

    1.When we implement elastic scale, can we really improve our performance and can this lead to saving costs on SQL databases?

    Absolutely.  If you look at the table below, you can see the ~Cost per DTU/month. For a P3 you are paying $4.65/DTU/month.  If you were to scale out, with the same number of DTUs on an S2, for example, you'd achieve an 67% savings on cost for the same number of DTUs (800).  Please keep in mind that there are feature differences between Standard and Premium SKUs as well as total size (250GB vs 500GB) - these may or may not affect your application.

    **Please note that S3 above is the preview price (50% off).

    2. Is there maybe a way we can easily test this by running queries or setting up an environment  where we can see real differences in DTU's per Shard?

    The available DTUs per shard is a constant value.  In principle, 800 DTUs on one P3 database is equivalent in performance capacity of 100 DTUs on eight P1s.  The test you want to perform is a comparison between your scale-up solution versus a scale-out solution as perceived by both the database %DTUs consumed and the response time/throughput of your client application. 



    • Edited by Joseph Idziorek Wednesday, February 11, 2015 4:02 PM
    • Proposed as answer by Joseph Idziorek Thursday, February 12, 2015 2:08 PM
    • Marked as answer by Elmar Maan Friday, February 13, 2015 10:15 AM
    Wednesday, February 11, 2015 3:59 PM