Notes from the Momentum 2011 session “Current and Future Architecture of Documentum”

These are notes from the session with Jeroen van Rotterdam, Chief Architect, IIG Services. It may contain errors and all these sessions are subject to change from an EMC perspective.

The focus on the Documentum 6.7 release was improved quality and performance improvements

Gives an example from a classic HA Configuration consisting of:


4 Web Servers

1 DocBroker

2 Content Server


He sometimes gets the question: ”Why is it so hard to deploy DCTM?” He smiled and exclaimed ”You guys want to do complicated stuff”.


The current components of the Content Server Repository:

–       Content Files (FS)

–       Metadata (RDBMS)

–       XML Store (xDB)

–       xPlore Full Text (xDB)

External sources

–       Centera

–       Atmos


Gives another example of a customer with 20k users

Branch Office Caching Server

–       Predictive caching (push content)

–       Distributed write option (async and sync). Local write and then syn cup.

The idea is to monitor users in a similar type of context.

Some users usually starts with an activity and will be in that process flow and therefore it is his/her context. Content related to that context can then be pushed to servers close to the user.



xMS is yet another acronym which in this case means xCelerated Management System

–       Define requirement – Blueprints

–       Describe them independent of deployment options

–       Automatically deploy blueprint to a target


In the Run component there can be:

-multiple VM Clusters running on

-multiple ESX

-virtual machines are created based on the blueprints and will be assigned ESX servers


The final component is what they call the Telemetry Project

-Monitor the runtime using open-source Hyperic

They have created hyperic adaptors to the Documentum products.

Integrated with the Integrien product (which now seem to be VMWare vCenter Operations)

Policy also includes upscaling configuration so it is easy to add more power to a configuration.

Automatic remedies like firing up an additional virtual machine

Total amount of metrics


Session optimizations

DFC Session Pooling

DFC frees session to pool if idle for 5 seconds

Expensive to switch context for users (to make sure they don’t see what the other users where doing)


Platform DFS Services/Platform Rest/Application Services



Two type of services

Core Platform and CMIS on top of that

Generate Application Services based on modeling from xCP stack (simple to use REST services will be generated for a specific part of the model)


Builder Tools:

–       Application Modeling

–       UI Builder

Semantic Model of the Application

Generate Optimized Runtime
– Indices etc

The Value of xCP is not just the UI but the application services and optimzed runtime is also of great value. Argues that xCP is sometimes misunderstood in that sense.


Dormant State Feature D7

Needed to support cloud deployment

No downtime

Bring the server to a dedicate state for changes (read-only, stopped audit trail, stopped indexing).

Partial availability for users in this state.

The idea is to spread update load on different content servers

Rolling upgrade – continues operation – apply patches on by one

Snapshot of the vApp is possible because it is in a safe state


NGIS – Public Cloud

Goal is full-blown multi-tenent architecture

Tremendous investment in xDB over the past years.

Argues that xPlore now beat search vendors FAST, Autonomy, Endeca and since all of them are bought by a big player EMC now has access to solid search technology.

Tenant level backup in xDb 10


–       XACML Security

–       Tree compression (previous version is stored as a change)

–       Search over history (storing complex graph that allow you to query all the versions)

–       Distributed Query Execution


Big Data becomes Big Information when you Put Smart on top of the data

Bring processing to the data rather than data to the processing

Impossible with the huge amounts of data of tomorrow to bring data to (central) processing nodes.


Plain Hadoop will not work in this case…plain MapReduce is optimzied for back-end.

We need real-time MapReduce processing a lot of research ongoing right now.

Stream-based (looking at Yahoo).


SmartContainers (next year)

Kazeon is integrated into NGIS

Offering a builder to model your metadata to generate the run-time

Early access program is available.

2 thoughts on “Notes from the Momentum 2011 session “Current and Future Architecture of Documentum”

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.