Capacity planning test results data

Windows Azure Pack capacity planning tests show the following results.

 

Instances of management portal for tenants

Requests per second

Response time*

Maximum number of concurrent users

Average CPU usage

Peak memory (Private Bytes) usage in MB

1

46

Less than 3 seconds

600

80%

950

2

90

Less than 3 seconds

1200

80%

1000

*The calculated page request response time does not include response times for client posts to the management portal for tenants, which involve performing tasks such as creating websites or creating end-user databases.

Throughput increased by 100 percent after increasing the instances of management portal for tenants from one to two. This happened because none of the downstream components (API and all the resource providers) hit bottlenecks. The management portal for tenants is expected to hit the CPU bottleneck before any of the downstream components.

Adding instances of the management portal for tenants increases the throughput linearly as long as the downstream components do not hit any bottlenecks.

 

Concurrent users

Average CPU usage

Component

Component CPU usage

Peak memory (Private Bytes) usage in MB

600

80%

Tenant API

10%

620

600

80%

Websites resource provider

7%

650

600

80%

Service Provider Foundation

6%

1360

600

80%

SQL Server resource provider

5%

420

600

80%

MySQL resource provider

3%

150

1200

80%

Tenant API

21%

640

1200

80%

Websites resource provider

7%

650

1200

80%

Service Provider Foundation

10%

1450

1200

80%

SQL Server resource provider

5%

550

1200

80%

MySQL resource provider

4%

160

The following table shows the tenant API performance results that use two resource provider instances.

 

Tenant API instances

Requests per second

Average CPU usage

Peak memory (Private Bytes) usage in MB

1

320

80%

1100

A single instance of the Tenant API was observed to reliably support 320 requests per second with an average CPU usage of 80 percent.

If you expect a high load on the tenant portal, it is important to monitor the resource usage on the machines. For information about monitoring performance over time easily, see Performance Monitor.

For information about stress testing a custom load, see Chapter 18 – Stress Testing Web Applications.

The SQL Server hosting server had 8 GB RAM and two processors. It was a stand-alone server and didn't use any type of storage area network (SAN) disk.

 

SQL Server resource provider instances

Requests per second

Average CPU usage

Peak memory (Private Bytes) usage in MB

1

26

21%

580

The best measure of performance for the SQL Server resource provider is the SQL Server operations disk I/O because disk I/O slows before maximum CPU usage is reached.

One instance of SQL Server resource provider can reliably process 26 requests per second with the response time of less than 3 seconds. Any load greater than this CPU usage degrades throughput, and response time increases to over three seconds.

Vis: