Performance Challenges on the Cloud

 

 

A typical cloud implementation could have hundreds of CPUs running in parallel. Some cloud solutions could be like an S-class Mercedes, however to leverage its muscles, it needs to be configured and fine tuned to reflect an efficient model. Examples of factors affecting a cloud implementation performance would be the number of Key Figures in an Integrated Business Planning (IBP) solution. The more Key Figures the higher the demand on the system and lower the performance.

 

Examples of common cloud performance challenges I have seen on SAP and Salesforce implementations are:

• Slow login time which is usually driven by network latency, demand on the enterprise SSO or other security system, firewall settings, or network device configurations.
• Dropped connections could be difficult to troubleshoot. On one implementation SAP users would receive a dropped connection error on an intermittent basis after about 30 minutes of being active on the system. Two weeks of troubleshooting uncovered the root cause of the issue was not due to the SAP cloud, but a misconfigured load balancer group of servers at one of the client's data centers.
• Initial load of templates, views and report runs are common issues when customers process large amounts of data. On a healthcare SFDC implementation the customer had a Contacts view for all its members, a whopping 1.5 million records. Obviously loading the view would take a few minutes of its agents staring at their screen. Educating the client on customizing the views to show members in a particular region, or who match certain filter criteria allowed them to view smaller data sets which were also more meaningful to their work.
• Refresh time is also a factor of the amount of data being retrieved, the browser
• Simulations
• Long running queries

 

Vendors constantly monitor their cloud operations checking memory utilization, background jobs status, CPU utilization, table sizes, system activities, and other performance KPIs to ensure optimum performance for their customers.

SAP key figure calculations can reach 10 million rows per second for its IBP solution in some implementations. Some factors that can affect that metric are data aggregations, projection, joins and uinions. In general loading a Planning View is dependent on the slowest Key Figure atomic calculation speed and the volume of data it processes. Using filters can sometimes be effective if it reduces the number of stored Key Figures that are defined as inputs to a calculation chain. However if filters do not affect the number of key figures and only affects data size the impact on performance might be minimal.

 

Are You Asking Your Engineer to Do the Impossible?

It is important for engineers building systems to understand how the organization will support the utility of their end product. I was asked to weigh in on a dispute between an IT vendor and a large multinational corporation. The vendor claims the IT security work has been delivered and they installed the systems and built the infrastructure needed. The IT sponsor at the corporation claims the implementation is not working.

 

Upon closer observation and after discussions with the technical team members I was able to identify the root cause of this confusion. Both entities were correct, the vendor indeed did deliver the work per technical spec, and indeed it was not serving the business objective the sponsor expected.


In a nutshell, the technical implementation was done correctly. However the way it was used by end users was not what both parties had envisioned causing certain portions of the solution to go out of sync with other components in the system, causing configuration errors yielding diluted value of the implementation. What was missing was an organizational policy that guides end users to follow certain process changes. Once the end users would follow the new organizational policy, and back doors were no longer allowed, the system components would be in sync and operate correctly.

 

The summary of this experience is that for a technology implementation to work properly the following need to occur in advance.

• Clearly articulate the updated operational process flow
• Ensure governance and policies support the new process flow
• Develop controls in the IT implementation to to enforce the policy

Without the three components the net realized value and user experience will be suboptimal. This is of course assuming all other standard best practices of system development have been followed from concept to deployment.