29-10-2012, 12:27 PM
Streaming and Virtual Hosted Desktop Study
Streaming.pdf (Size: 336.21 KB / Downloads: 24)
ABSTRACT
As part of an ongoing evaluation of emerging computing models for enterprise operations,
Intel Information Technology studied the impacts of streamed client and Virtual Hosted
Desktop computing models on server and network utilization. Utilization varied depending
on the model studied, illustrating that each model is more efficient for specific workloads.
Virtual Hosted Desktop is better suited for standardized applications with relatively static
screens, such as data entry. Streaming more efficiently supports multimedia and other
dynamic applications, including unified communications.
Background
New computing models are emerging in the enterprise.
As networks become more robust, the common theme is
to enable centralized, network-based services that are
provided on-demand.
One of the computing models that meets these criteria is
streamed client computing. With streaming, the server delivers
the OS and/or applications over the network for temporary, local
execution by clients. OS streaming involves creating and storing
a disk image on a server and loading it on the client via the
network at boot time. In application streaming, the server stores
strategically packaged client applications and distributes them to
the clients as users access the application. When a user invokes
an application, the server sends the first application execution
block to the client, allowing the user to immediately become
productive. It sends additional blocks in the background.
Streaming can work with virtualization technologies. For example,
streamed applications can execute within a Virtual Machine
(VM) on a client or server.
Benchmark Study
To evaluate the impact of these computing models, Intel
Information Technology constructed a performance study
to characterize backend utilization under a typical user workload.
The load included standard office suite applications, plus
concurrent playing of a video in a media player. We focused
on the impact to backend resources by capturing server and
network metrics as the load was scaled from one to 20
simultaneous clients.
Conclusion
Based on our study of streaming and Virtual Hosted Desktop computing
models, it is apparent that each computing model is efficient
for certain workloads, but also requires special considerations when
designing around a particular model. Table 3 summarizes our results.
Virtual Hosted Desktop offers excellent performance for certain
workloads, allowing infrastructures to run as much as 20 times
thinner. However, Virtual Hosted Desktop, also places a substantial
load on the server, up to 45 percent utilization for 20 clients. Scaled
out further than 20 clients, we would expect to run out of server
processor resources before we flooded our network capacity. The
study shows the importance of system design for Virtual Hosted
Desktop to prevent overloading the server especially, if multimedia
and other demanding applications will be used.