VMWare ArcGIS Server Deployment Guide Available
VMWare and ESRI have jointly published a deployment guide for ArcGIS Server 9.3.
An excerpt from the introduction:
This document describes the best practices for running a typical ESRI ArcGIS® Server deployment on VMware® Infrastructure. It provides basic guidance on the architectural design of an ArcGIS deployment and the value of utilizing the VMware platform to provide virtual infrastructure. It then discusses results of recent testing conducted jointly by VMware and ESRI characterizing the performance and functionality of ArcGIS Server running on VMware infrastructure version 3.5i update 3. Finally, it outlines some best practices for utilizing the two products together in your datacenter.
The guide provides benchmarks for both physical and virtual machine configurations as follows:
From: "ESRI ArcGIS Server 9.3 Deployment on VMWare Infrastructure"
From: "ESRI ArcGIS Server 9.3 Deployment on VMWare Infrastructure"
From: "ESRI ARCGIS SERVER 9.3 DEPLOYMENT ON VMWARE INFRASTRUCTURE"
From Table 1 and Table 2 we can see that physical servers significantly outperform virtual servers with equivalent hardware resources. The only apples-to-apples configuration provided is for 4 cores, 16GB RAM where all ArcGIS Server components were configured on a single server. In this configuration the physical server had a throughout of 42,264 transactions/hr and the virtual server had 27,432 transactions/hr, a performance decrease of 35% on the virtual server.
Of course some performance decrease is to be expected and that is a trade-off with the benefits of virtualization over physical hardware (availability, manageability, ease of provisioning etc.).
What is interesting to note is how well ArcGIS Server can be scaled on VMWare (Table 3) by increasing the number of SOC machines and by using network attached storage. With 1 SOM and 7 SOCS (total of 8 cores and 16GB or RAM) the throughput was measured to be 52,344 transactions/hr which is very similar to the performance of a single 8 core, 16GB physical server that had 56,574 transactions/hr (a 7% decrease in performance). It should be noted here that the VMWare benchmark was conducted with RAID-5 network attached storage while the physical server benchmark was conducted with local RAID-1 disk.
See the full deployment guide for specific details and recommendations.
