proposed bench test for gis servers
DESCRIPTION
ArcGIS Server, WiDNR, performance, capacity, benchmarkTRANSCRIPT
Our business needs
Replace ArcIMS applications with ArcGIS Server
Replace 16 existing W2K3/AMD 275 blades with ?
We chose to focus on throughput and performance
We have, by some definitions:– 39 applications– 14 million hits / month– 40,000 users / month
We knew ArcGIS Server was resource demanding
We knew ArcGIS Server could be slow compared to ArcIMS
How to define “capacity”
We defined capacity as the maximum number of map requests a host or system can serve per second (“maps / second”)
How to define performance
We defined “performance” as the average time it takes a server to respond to a map request (milliseconds or seconds)
Target parameters
We estimated that our ArcIMS environment was capable of producing about 40 maps / second
We estimated that our average ArcIMS response time was 1 second (1000 ms)
Load testing client
JavaScript application (use browser as client)
“export_map” requests to REST API
Configurable parameters– Host– Map services– Number of asynchronous
requests
Randomizable parameters– Map services– extent
Map service
Main map service for one of our major, planned AGS applications
Large number of layers Geometrically complex layers Contains imagery
SPEC
Standard Performance Evaluation Corporation
www.spec.org Hardware providers conduct standardized
tests and report results Spec_int_rate2006
Test map requests—why?
Our tests have shown that for ArcGIS Server, at least, mapping remains the most resource intensive operation
OPINION: I suspect it is why people come to your site and that almost every request of another type will involve mapping
Proposed vector map service
Multiple data layers Nationwide coverage Variety of scales Geometrically complex Relatively uniform complexity across the
country
Proposed image map service--why
In ArcGIS Server, at least, rendering dynamic imagery remains most expensive operation.
Can choose to avoid this by using cached imagery Make image test separate
Client factors
Web client Client can have significant effect on performance Randomizable / separate tests?