Right sizing and performance - Using Ellexus Breeze for EDA Workload Migration to AWS

Right sizing and performance

EDA tools are known for being CPU and I/O intensive. Figure 5 shows the I/O patterns, CPU, and memory usage for the EDA tool. At different times of the synthesis run, the EDA tool is I/O, CPU, and memory intensive. Applications that perform many small read operations, as this one does, are often both CPU and I/O bound as they are limited both by the storage latency and the speed of the CPU to issue the I/O requests. This is supported by the CPU pattern, which hovers between 200% and 400%, while the number of I/O operations is about 80K IOPS.

This data implies that the AWS Cloud infrastructure should have several fast CPU cores and a storage solution that can perform at 80K IOPS.

Figure 5: Breeze Timeline view shows the I/O, CPU and memory of the EDA tool over time

The Breeze Profile view shows I/O, CPU, and memory usage over time. This view includes a detailed breakdown of storage performance and the impact that has on the workflow. Figure 6, taken from the Profile view, shows how much time is spent in read operations per second. This workflow can spend over 500ms per second waiting on read calls. It’s likely that the workload is multi-threaded so that many operations can be issued in parallel, and could benefit from having fast local storage. Some experimentation is recommended to determine which storage option gives you the best return on investment.

Figure 6: Read latency over time, taken from the Breeze Profile view