The primary virtualization technologies utilized by ESRI customers are:
- Application Virtualization
- Hardware Virtualization
Many customers have been utilizing Application virtualization solutions to assist with distributed rich client applications and a centralized datasets via products such as Microsoft Terminal Services and Citrix Xen App. The primary performance advantage is that these solutions do not need to send large amount of GIS data to the clients, only light-weight screen refreshes.
Recent, hardware virtualization has been increasing in popularity. Hardware virtualization solutions are being broadly used in test, development, and production environments, providing tremendous benefits to ArcGIS Server deployments. ESRI uses virtualization technologies for the development, quality assurance, and certification processes of ArcGIS Server.
Running software in a virtualized environment affects the performance of any application to some degree. Like any application, ArcGIS Server performance is negatively impacted by virtualization. As the workload on server processing becomes heavier, the decline in performance becomes more obvious.
Virtualization is a fast and efficient way to perform server consolidation within a data center that has many machines that are underutilized. The key concept to remember when considering this technology is, “will consolidation help within the data center?” and if so what parts can be consolidated. Some key indicators to keep in mind when making the decision to virtualize include how many servers are running under capacity and what is the projected user load for each server.
Performance Factors
Virtualization tests performed by ESRI indicate that intensive disk I/O operations, such as dynamic mapping and map caching, perform faster on physical machines than on virtual machines. Some CPU-intensive applications are also negatively affected in a virtual environment. Testing has shown that the performance can differ by virtualization vendor (sometimes dramatically) and by the operation being executed. For example, calls to ExportMapImage on a moderately complex map service (40 vector layers and 3 raster layers) through our REST API translate into a 20 percent to 30 percent performance hit, depending on the virtualization vendor. It is important to highlight that a suboptimal configuration of the virtual environment can lead to larger performance hits of up to 60%.
Tools
Related Links
- VMWare performance best practices and benchmarking tips
- Microsoft Hyper-V
- Citrix XenServer
- Amazon EC2
For more information
- ArcGIS Server and Virtualization Whitepaper - Hardware Virtualization guidance from ESRI
- Virtualization Best Practices - Virtualization recommendations from Computer Associates (CA)