Is there any rule of thumb on the number of caching services to have with Content delivery servers. Like 2 Cache services with 1 CD in Azure PAAS.

Asking this question, because one of our clients faces 100% CPU usage on Caching service and Content delivery services, for at least 15-20 minutes after the Sitecore Azure PAAS build completes successfully and web role is started. This issue we are facing after the Client added 2 more Content delivery servers (in anticipation of more site visits) during the festival week in China.

So do we have to increase the Caching service count from current 2 to 4? The only issue is high CPU usage after the build and sites are unavailable for at least 25 minutes after.

Currently, we have 4 CD's but only 2 cache services. I understand that the number of Caching services may be dependent on the No of Site hits and other factors as well. Does Azure support team help clients in issues like these to verify Site installations?

  • Would you mind changing the question title to something a little more specific to the issue you're seeing? I thought about changing it myself, but I'm just a tad unclear on exactly your issue. I don't want to incorrectly change the title. Oct 28, 2016 at 5:26

2 Answers 2


My reply does not answer your question directly, because Azure caching is to be defined dynamically according to your load factor.

Nevertheless, for the high load website I would personally recommend the following:

  1. Tune your Sitecore rendering output caches as per Sitecore official performance tuning guide (your case is section 4.4, but it always good to get yourself acquainted with all the information) https://sdn.sitecore.net/upload/sitecore7/70/cms_tuning_guide_sc70-72-a4.pdf ;

  2. Do not deploy your code to all the servers at once. Instead, deploy one by one, to make sure there is at least one alive cluster

  3. Create a warmup script which will browse through the website sitemap and fill the sitecore output cache and azure cache for the CD instance that has got an updated code.

I don't have any figures (like % of reduced load), but correctly configured output caching always improves the performance dramatically.

p.s. As a general suggestion, I would also recommend you to use the dotTrace profiler to identify places in code, which utilize most of the CPU time and cause the load. This will help you a lot to identify places in your code to optimize and debug.

  • Hey thanks @adb for the suggestions. Actually the client has two sites in Azure PAAS and IAAS , the latter one is specifically for China(govt laws) and have the same code base. The high CPU usage is seen only on Azure PAAS.
    – Abhay Dhar
    Oct 28, 2016 at 5:58
  • @ABDmaverick then my answer is irrelevant at all. I will keep it here for people's future reference. sorry for that Oct 28, 2016 at 7:31
  • No, the answer is not irrelevant . Infact, it does help me in understanding Azure ( on which I have very less experience), For a build done through Sitecore azure module , Can you tell us me, if its possible to do deployments in incremental way ( As per point 2 of your answer)
    – Abhay Dhar
    Oct 28, 2016 at 7:39
  • @ABDmaverick I never used the Sitecore Azure module for deployments, sorry Nov 2, 2016 at 3:27

Microsoft is going to retire Azure In-Role Cache on 30-Nov-2016, which Sitecore Azure module uses as a storage for session data. It's recommended upgrading the module to the latest version of 8.0 rev. 161110 or 8.1 rev.161109 and start utilizing the Azure Redis Cache with Sitecore Redis Session State provider.

For more details, see the following article:

  • Sorry just read the release notes , session end is supported . gr8 news!! Thank you oleg
    – Abhay Dhar
    Nov 16, 2016 at 11:12
  • On following the Sitecore document on upgrade , I cant get the connection string for Redis to get to work
    – Abhay Dhar
    Nov 30, 2016 at 6:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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