Validating memory access traces

[Suyash Mahar]

Hi all,

I am using DynamoRIO to attach to workloads and collect memory access traces. However, I am trying to understand how to ensure that these traces are representative of the application's behavior, given the slowdown that DynamoRIO introduces which can affect machine-level scheduling or global load-balancers.

The simplest approach I can think of is to compare the read-to-write ratio from the collected trace to what perf reports without DynamoRIO. Are there better methods that people are aware of or have been using? 

I noticed that Google open sources workload traces from production workloads in steady state. Is there any information available on how those traces are validated?

-Suyash

[Kai Luk]

Perf can also be use to gather instruction samples without DynamoRIO and compare them to instructions captured by DynamoRIO to ensure they are running the same code. 

Kai