How Big Can CMS Infrastructure Get?

How big can a production CMS server get?

The short answer is as big as your blank check allows.

serverInfra

As you add users, workflow, translation, heavy publishing, and lots of opening/changing components, the CMS server size will inevitably increase.

Often times, multiple CM services are all enabled on one large machine for the enterprise, but the downsides of that are when one problem happens, all aspects of your CMS fail with it. Below I show what an infrastructure looks like that handles very large publishing loads and is scalable to add more load. The CMS also handles workflow services and translation integration with World Server.

You will have to click the image to open in a new window and zoom in to read the details.

Tridion PROD Infrastructure 2013

 

Turning on only the publishing and transport services allows you to move the rendering of pages (the publishing load causing performance hits seen to the CMS user) over to another server or pool of servers. This also allows your dedicated publishing servers to scale as publishing load increases. This means that if publishing isn’t keeping up with demand, you can spin up another machine and add it to the pool to gain throughput. The less time editors stare at “Waiting for publish” the more work they complete!

The advantage of doing this doesn’t stop at scaling. Now your users do not see a slowdown in system performance when publishing. If publishing needs restarted, then the users can still work in the CMS and publishing resumes when the server comes back up. The publishers can be restarted in series to allow for zero downtimes.

In addition to the dedicated CMS publishers, you can also repurpose a dedicated DR (Disaster Recovery) CMS box to also pick up additional capacity when not in use. If you need the DR server it is as simple as starting the rest of the services and it is back fully purposed as a DR CMS environment in minutes.

As for the DR environment, it’s a fairly simple setup. Log shipping is setup between the Live CM database and the DR CM database. If DR environment is needed, you can simply turn off the log shipping and the DR CMS will start working.

There are some additional ways to scale out as needed. If you have a lot of workflow processes running you can have a dedicated machine just to run your workflow processes. This has similar pros to the publishers.

All static files go to central data center NAS servers which serve the individual web servers with the static content.

The size of the CMS machines are fairly large machines in the Tridion world. For production, we have 3 dedicated publishing servers with 48 CPUs and 256GB memory each. This allows for an optimal 108 rendering threads and 72 deployer threads for high-speed mass publishing.

 

 

 

 

Problems with Container Based Content Model

Containers are a common approach when designing CMS content models. They allow Authors and Content Editors to organize content hierarchically. While containers allow contributors to group, reorder, and organize content easily, they also add a lot of dependencies to the content across the system.

Containers add complex scenarios for simple tasks, including:

  • Publishing
  • Delete + Unpublish + Localize + Unlocalize
  • Workflow
  • Translation

Effects of using a container model:

  • “Where Used” filters are used extensively
  • Large amount of content dependencies lead to performance issues
  • Large amount of content manager transactions while publishing to ensure content integrity lead to stability issues
  • Multiple Database Transactions during Publishing
  • Complex content unpublish and deletion
  • Complex Experience Manager implementations (Experience Manager is not designed to use Containers)

Content Integrity Example:

In the scenario below, Tridion will attempt to maintain content integrity by publishing all pages using component instances. This example would cause a link propagation issue if the following conditions were met:

  1. Text Block – With image component is published within a page
  2. Application Include Container component is published within a page
  3. Buttons Actions tray component is published within a page
  4. Button App Include is published within a page

containerImage

Web 8, SI4T and Storage Extensions

The release of SDL Web 8 has seen the biggest overhaul of the Content Delivery stack since Tridion 2011 and much of it has been applauded much, as the new Micro Services architecture and with it the ability to scale, deploy and be distributed in general, gives many more architectural options than just having an API in your web application connecting you to a database.

Building this while guaranteeing backward compatibility with the Tridion 2011/2013 CD stack is more than a major task and I for one am happy to see that that feat was pulled off really well. That is, unless you have overridden the default DAOFactory classes in your Deployer Storage Extensions, as is the case with the SI4T Storage Extensions. When testing SI4T with Web 8, it turned out that loading of the SI4T Storage Extension did not work in every scenario. This has to do with the subtler points of Spring Bean Class loading.

Continue reading

Caching In: Revisiting Data Sharing across Templates

A while back I wrote a post on using Context Variables as a templating cache. When I came across some requirements to cache data between templates on a recent project I revisited this concept, but found some limitations with Context Variables which I had not realized before. As such I ended up implementing a different approach to caching

Continue reading

Building and automating a scaled out Fredhopper/SmartTarget environment

While designing and building a Fredhopper/SmartTarget enterprise environment recently, a couple of interesting requirements came up. The first requirement, these days quite often asked for when building infrastructures, was that every Fredhopper component needed to be to be automatically deployed, configured and run. The second hard requirement was that in a production environment being under constant high load and being distributed across multiple data centres, the Fredhopper Index servers need to be in sync and highly available in each data centre, while having failover mechanisms causing the least amount of disruption time. After a lot of headache, trialling, erring and creating a mountain of broken Fredhopper instances in the process, we finally managed to meet requirements and this post shows the how of it.
Continue reading

SmartTarget Extensions: Better Security on Promotion Management

If you have used SmartTarget, you will probably know that the security model for managing promotions is pretty basic – you either have rights to manage promotions or you don’t. On my current implementation we have different business units managing their own SmartTarget promotions for various sites and applications, all within the same CMS, so all able to view, edit (and accidentally break!) each others promotions. This post shows a simple solution we came up with to ensure each user could only manage appropriate promotions.

Continue reading

Running a .Net DD4T application on Linux

DD4T .NET on Linux

 

At the Tridion Developer Summit in September Siawash Shibana and Albert Romkes gave a presentation of a DD4T .NET application running under vNext (The codename for the next .NET framework) on Linux.

Siawash and Albert made the application available publicy and although it currently uses mock SDL Web 8 provider objects plugging this into a real life content delivery service should be really simple when SDL Web 8 is released this month.

Continue reading

Custom Renderer – Index Keyword Key in SmartTarget

keywords in SmartTargetI found out how to use a new CM-side extension point today; Renderers. I had been struggling for a while with the fact that SmartTarget does not index the Keyword key – which was precisely what I wanted to set up some promotion content selection filters on. It turns out to be quite easy, and indeed gives a generic method for messing with the content that SmartTarget indexes without having to fiddle about on the Delivery side with Java or Kettle.

Continue reading