Assured Guaranty High Performance MatLab Risk Modeling on AWS

Brian McCallion Case Study, Financial Markets


Assured Guaranty insures municipal bonds against default and helps investors to mitigate risk and to provide liquidity to financial markets.


In quantitative finance modeling risk is an ongoing activity. Yet such models require large memory and CPU. In order to be able to analyze large datasets of terabytes of data such models consume large memory and CPU. In the data center developers often found the models took many hours to complete. This limited the size and number of models they could run in a single day. For the IT team the virtual machines had to be allocated large amounts of CPU and memory, yet the resources were used for periods of time, then not used at all. Both the developers and the Operations Team sought a better way.


By creating a Windows Amazon Machine Image Bronze Drum enabled developers to launch MatLab workloads on demand in the Cloud. To enable high availability of model and scenario data, Bronze Drum created a process that periodically issues an s3 sync command to sync the EBS volume of the workstation with S3 data. As a secondary measure, an AWS Lambda function periodically executes a snapshot of the MatLab boot and root volumes. To manage costs a Lambda function starts the instances in the morning, and shuts them down at the end of the day.  AWS EBS backed instance types, can be stopped, the instance type changed to a higher performance instance type, and the instance started up again. In other words, the instance type can be changed with a simple Stop and Start, and so developers can quickly switch to a high performance instance and get the result in minutes rather than hours.


Time to Market

The firm saves money on both scarce developer time, while removing compute resources as a constraint. The mission critical risk modeling activity can be done using the most appropriate resources and more and larger models run faster and at lower cost.


Developers use smaller instance types while coding their model, then simply stop the instance, change the instance type with large amounts of CPU and memory. Based on the type of workload developers can choose from the C4, X1, or even the G2/P2 instance types and run the model to completion in far less time than in the data center. The IT is able to dedicate fewer resources.

Cost Savings

As a result of this initiative developers are more productive and the converged storage and memory they were consuming is replaced with cost efficient S3 and EBS. Developers are able to obtain very large amounts of memory, even leveraging X1 spot instances by being strategic about when they run the workloads, or simply run large instances for a short period of time so as to complete the modeling fast.

About Bronze Drum Consulting

We bring high octane technical skills and automation at price points that enable us to work with firms in the early stages of development, strategy, and delivery. We help executives manage engagements with global consulting firms, advise business units, and deliver world class applications. Our deep investment in AWS Training means that we deliver solutions that align with Cloud Best Practices and continue to optimize beyond initial deployment. We engage early and deliver long term value.