Database Experimentation Assistant – FAQ


Distributed Replay Recommendations

1. How much additional hard drive space is required on Dreplay Controller machine?
Given the production trace, additional 78% hard drive space is needed to create IRF files on the controller machine

2. What is the ideal trace rollover size used for capturing production and/or performance traces?
200 MB or 512 MB

3. What is the expected Replay Time given captured time of a production trace?
Replay time is approximately 1.55 times the capture time (given that we use one controller, 4 children for replaying the production trace).

4. What is the minimum CPU and RAM requirement for Dreplay controller and client machines
Single Core CPU, 3.5 GB RAM at the very least.

5. What is the size of Performance Trace on Target Server given ProductionTrace?
Assuming that our “published” versions of production and performance trace definition files are used, and the performance trace definition filters the traces out for one database of interest: our experiments have shown that

Performance Trace = 15 * Production Trace (approximately).

Database Experimentation Assistant Tool Recommendations

1. What is the minimum hardware requirement for running the tool?
Single Core 3.5 GB RAM machines

2. What is the ideal hardware requirement for running the tool?
8 Core CPU (with RAM 3.5 GB or higher). More than 8 cores does not help DEA runtimes.

3. What is the recommended hard drive requirement for running the tool?
Additional 33% of performance trace size is needed to store A, B and Analysis databases.

Other

1. Does the tool automation clean up trace files created?
No, the tool does not clean up any trace files in the folders you used on the machines. You are required to manually clean up the files created by tool.

2. What are the error codes shown in the tool?
Errors shown by the tool are of severity >=10. This page explains the severity levels of errors raised by the SQL Server Database Engine. For a complete list of Error Codes, please find detailed error information from the SQL Server technical documentation library. Under this page is the technical documentation for all supported versions of SQL Server.

3. In Analysis report, what does the “indeterminate” queries mean?
Indeterminate queries are classified when we do not have the sample size to perform a statistical analysis on the query. Currently, the tool requires that a query is run at least 30 times on A and B in order for statistical evaluation to be performed.

4. Why are the error counts different between A & B instances?
In certain situations, the trace events may be flushed out of order to the trace file. If the error event (User Error Message) is flushed before the batch starting event (SQL:BatchStarting), the error is not associated to any batch. In this scenario, when the same workload is replayed on A and B instances you might see that the execution count is different for the two runs.

5. What SSDT version should be installed?
For Analysis, SSDT versioning should be same version as the Microsoft SQL Server.

Feedback/Contacts

If you have any questions regarding the execution steps outlined in the sections or run into any issues with the tool, please email: deafeedback@microsoft.com

Comments (0)

Skip to main content