Microsoft Office Tutorials and References
In Depth Information
Optimizing for Performance
5.
When rendering the page, the browser makes another call to the WFE for the chart image.
6.
The WFE calls the ECS to generate the chart image.
7.
The ECS generates the chart image, stores it on the local disk cache for future use, and returns it
to the WFE. The ECS uses the Graphics Device Interface (GDI) to render the chart image. This
operation allows only one chart to be generated at any time in an ECS process.
8.
The WFE return the image to the browser, which displays it.
Charts can be rendered only through the EWA. The API does not support rendering charts.
Optimizing for Performance
This section analyzes the various architectural aspects of Excel Services from the performance point of
view. You expect good performance and full utilization of your hardware. There usually is a trade-off
between the response time and throughput, and the other aspects of the server such as security, data
staleness, and functionality. Excel Services has a large number of ways to help the administrator and the
author reach the balance that is best for them.
There are two major measurements for the performance of the server: response time and throughput.
Response time expresses how the end user measures the performance of the server from the moment
he or she performs an operation until the results come back and are displayed in the client. Throughput
measures the capability of the system to perform a number of parallel requests from multiple users. It is
measured in terms of requests per second that are actually performed.
The performance of the server depends on the optimal utilization of its resources, including load
balancing, network, I/O, memory, and CPU. The following sections explain the various factors that impact the
usage of each of these resources, and provide tips on achieving the right balance between performance
and functionality.
Load Balancing
Excel Services supports the following three ways of load balancing the ECS machines when a new session
is started:
Local load balancing — The WFE opens the session on the ECS that exists on the same machine. Use
this load balancing when each machine contains both the WFE and the ECS, and the expected
request mix is mainly opening diverse workbooks. For this scenario, caching on the ECS is not
relevant, so working with a local ECS ensures the most optimal load balancing by the NLB on
the WFE, and no network communications is needed between the WFE and the ECS. On the
other hand, if sessions are expected to have multiple requests, you should not use this
configuration, because the requests after the first one might be served on a WFE that is located on a
different machine than the ECS.
Round-robin — The WFE picks a random ECS machine to open the session. This load balancing is
best if there are a few workbooks (so they can all fit in the memory of each ECS) and they are not
used uniformly. This method distributes the requests evenly, achieving good utilization of the
Search JabSto ::




Custom Search