Upload
vuongnguyet
View
248
Download
2
Embed Size (px)
Citation preview
ERDAS APOLLO PERFORMANCE BENCHMARK ECW DELIVERY PERFORMANCE OF ERDAS APOLLO VERSUS ESRI
® ARCGIS FOR SERVER
White Paper April 14, 2014
ERDAS APOLLO Performance Benchmark
April 14, 2014 2
Contents
Introduction .............................................................................................................................. 3
Sample Dataset ................................................................................................................... 4
Test Hardware ..................................................................................................................... 5
Summary ................................................................................................................................... 6
Methodology ............................................................................................................................. 7
Server Configuration ........................................................................................................... 7
Benchmark Configuration .................................................................................................... 9
Results ...................................................................................................................................... 12
Response Time ................................................................................................................... 12
Hardware Performance ....................................................................................................... 13
ArcGIS for Server during load test ................................................................................. 13
APOLLO Essentials during load test ............................................................................. 14
Data Transfer ...................................................................................................................... 15
Response Time Distribution ................................................................................................ 15
Response time vs Users ..................................................................................................... 16
ERDAS APOLLO Performance Benchmark
April 14, 2014 3
Introduction This whitepaper aims to contrast the performance characteristics serving the Enhanced Compressed Wavelet (ECW) file format from ERDAS APOLLO against Esri® ArcGIS for Server. Even when the storage format and hardware is identical, there are substantial differences in how geospatial server applications deliver this data from a performance point of view. Critical enterprise deployment questions include:
Response time under low, consistent user load
Response time under increasing user load
Response time consistency
Response time across various view sizes
CPU & memory hardware utilization
Bandwidth requirements
Cost effectiveness
All these concepts are addressed in the following sections by performance testing the OGC Web Map Service (WMS) interface common to both server platforms.
ERDAS APOLLO Performance Benchmark
April 14, 2014 4
Sample Dataset
Australia_Topographic_Map.ecw
File type: ECW Image
Dimensions: 146,105 x 122,280 px
Gigapixels: 18
Projection: EPSG:4326
Cell size: 0.000286 degrees
Structure: 3 Band, RGB UINT8
File size: 1,999,168 KB
Figure 1 Dataset extents
Figure 2 Native resolution sample
ERDAS APOLLO Performance Benchmark
April 14, 2014 5
Test Hardware All performance metrics found in this whitepaper were performed on the following hardware.
Data drive throughput, reported by ATTO Disk Benchmark (4QD)
Warning: The intent of this whitepaper is to highlight peak performance of ERDAS APOLLO on given hardware, under specific, selected test plans. It should only be used as a relative guide when estimating any deployment, as characteristics will vary.
ERDAS APOLLO Performance Benchmark
April 14, 2014 6
Summary
ERDAS APOLLO 2014 is 3.05x faster and more consistent at delivering the ECW file format than Esri® ArcGIS for Server 10.2 with Hexagon Geospatial’s “ECW for ArcGIS Server” installed.
50.34
2.54
170.33
150.57
5.96
649.01
0 100 200 300 400 500 600 700
600
2048
256
Requests / Second
Request size (px)
ERDAS APOLLO Essen:als 2014 vs
Esri ArcGIS for Server 10.2
APOLLO Essen5als
ArcGIS for Server
ERDAS APOLLO Performance Benchmark
April 14, 2014 7
Methodology
Server Configuration 1. Configure entry level 4-core test server, “PALE” with the latest software releases
a. ArcGIS for Server 10.2
i. With ECW for ArcGIS Server 2014
b. APOLLO Essentials 2014
2. PALE represents:
a. Windows Server 2008 R2
b. IIS 7.5
3. Configure test dataset for ArcGIS for Server
a. Add ECW data to ArcGIS for Desktop, add ECW > Publish to Server
b. Enable WMS service capability
i. http://pale:6080/arcgis/services/largeECW/MapServer/WMSServer?
Figure 3 - Publish ECW to ArcGIS for Server from ArcGIS for Desktop
4. Tune ArcGIS for Server to obtain highest throughput possible
a. Test case proved the highlighted configuration to be preferred
i. High isolation, 1->2 instances = 121ms average (default setting)
ii. Low isolation, 1 instance, 8 processes = 157ms average
iii. High isolation, 1->4 instances = 79ms average
ERDAS APOLLO Performance Benchmark
April 14, 2014 8
5. Tune ECW for ArcGIS Server ECW cache
a. A value of 10% yielded the highest throughput
i. <SDKCacheSize>0.01</SDKCacheSize>
b. 10% of 24GB total system memory adds ~240MB to each ArcSOC.exe. So we can estimate under load, each ArcSOC should not exceed 500MB
6. Configure test dataset for ERDAS APOLLO Essentials
a. Connect to Server > Add Image
b. Enable WMS Protocol > Press apply
i. http://pale:20140/erdas‐iws/ogc/wms?
ERDAS APOLLO Performance Benchmark
April 14, 2014 9
Figure 4 – ERDAS APOLLO Essentials administration console
Benchmark Configuration 1. Three different test cases were identified.
a. Random 600 x 600px WMS views with stepped user load up to 20 users. Time bound to 6 minutes
b. Random 2048 x 2048px WMS views with consistent 2 user load. Request bound to 1000 requests
c. Random 256 x 256px WMS views with consistent 10 user load. Request bound to
2. Generate random WMS extents across the dataset, maintaining aspect ratio and using all scales via the RandomWMS command line tool. Save output to CSV for use within test plans
ERDAS APOLLO Performance Benchmark
April 14, 2014 10
Figure 5 - RandomWMS generating extents for 256 image size
3. Configure JMeter test plan to capture the three tests. See appendix for full JMX
a. The stepping thread group used in the 600px test is defined in increments of 2 users, up to max of 20.
b. Basic thread groups for the other two tests, with 2 threads and 10 threads respectively
c. Configure HTTP Request Samplers to point to the relevant WMS end points to issue GetMap requests
i. Using BBOX defined in the CSV Variables $(BOX)
d. Configure relevant listeners to measure throughput and hardware utilization
ERDAS APOLLO Performance Benchmark
April 14, 2014 11
4. Execute JMX test plan from remote load generator machine, connected via Gigabit Ethernet to the PALE server
ERDAS APOLLO Performance Benchmark
April 14, 2014 12
Results
Overview
Label Requests Avg Min Max Std Dev Errors Throughput
(req/sec) KB/sec
Avg. Bytes
ArcGIS Server
600px 18126 229 33 845 125.3 0 50.34 5,792 117,815
ArcGIS Server
2048px 1000 785 262 1681 322.7 0 2.53 3,561 1,438,220
ArcGIS Server
256px 5000 53 15 175 18.8 0 170.33 3,439 20,675
APOLLO Essentials
600px 54208 75 12 819 42.3 0 150.57 12,283 83,536
APOLLO Essentials
2048px 1000 333 123 1285 159.1 0 5.96 5,729 983,796
APOLLO Essentials
256px 5000 10 3 199 7.5 0 649.01 9,444 14,901
Response Time
ERDAS APOLLO Performance Benchmark
April 14, 2014 13
Hardware Performance
Both servers successfully met the expected 100% CPU utilization at 4 concurrent users
ERDAS APOLLO Essentials substantial 256px test throughput can be clearly seen by the Disk IO counters
ArcGIS for Server during load test
Figure 6 ArcGIS for Server during load test with 4 processes
ERDAS APOLLO Performance Benchmark
April 14, 2014 14
ERDAS APOLLO Essentials during load test
Figure 7 – ERDAS APOLLO Essentials during load test with 1 process
ERDAS APOLLO Performance Benchmark
April 14, 2014 15
Data Transfer
Response Time Distribution
As highlighted by the result summary, ERDAS APOLLO Essentials shares a far smaller standard deviation which is clearly seen on the response time distribution chart above. Consistency is very important for baselining and defining performance criteria to more accurately plan and scale server deployments
ERDAS APOLLO Performance Benchmark
April 14, 2014 16
ArcGIS Server response time increases sharply once CPU Utilization reaches 100% at 8 concurrent users. While expected, the rate at which performance declines is far worse than ERDAS APOLLO Essentials which maintains a relatively linear, consistent trend ensuring more reliable performance can be assured under heavy load.
Conclusion The substantial 300% performance advantage seen with ERDAS APOLLO can be attributed to:
1. ERDAS APOLLO architecture
ERDAS APOLLO Essentials is a heavily optimized and multithreaded server. Unlike ArcGIS for Server that requires multiple processes to scale across CPU resources, ERDAS APOLLO utilizes just a single multi-threaded process to scale from 4 cores up to 32 cores which yields significant advantages including:
Higher memory cache hits
Lower memory requirements
Lower disk I/O requirements
No process duplication
Shared resources across map services
2. ERDAS APOLLO hardware acceleration
ERDAS APOLLO leverages additional CPU SIMD hardware acceleration above what’s available within the core ECW JP2 SDK shared by both servers. These optimizations ensure ERDAS APOLLO can generate imagery quicker than ArcGIS for Server, or indeed any other image server platform, across the whole image processing pipeline. This includes JPEG generation, to reprojection warping and resampling.
0
50
100
150
200
250
300
350
400
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Response :mes (m
s)
Users
Average Response Time vs Users
ArcGIS Server
600px
APOLLO
Essen5als
600px
ERDAS APOLLO Performance Benchmark
April 14, 2014 17
3. ERDAS APOLLO Native C++ application
ERDAS APOLLO is a native C++ application that provides direct communication with the IIS ISAPI handler on Windows, or Apache FastCGI on Linux. This direct access path removes unnecessary latency or complexity that is typically required within other server applications. For example within ArcGIS for Server, where there are multiple bridges between Web Adaptors, Java application servers, inter-process communications and process management required to manage child processes (ArcSOC.exe).
4. Hexagon Geospatial are the image compression and serving experts
For over 15 years the ECW format and what is now the ERDAS APOLLO server platform have gone through continual improvement, and we constantly leverage our experience in this area to refine these solutions even further. Because we design and build both, we understand exactly how to architect a high performance image server while leveraging the core capabilities of the file formats including ECW.
Despite the performance difference, the availability of ECW for ArcGIS Server continues to address the demand from existing users of ArcGIS for Server. While it will not represent the full performance capabilities of the format, it still offers improved performance for many Hexagon Geospatial customers.
ERDAS APOLLO Performance Benchmark
April 14, 2014 18
About Hexagon Geospatial Hexagon Geospatial helps you make sense of the dynamically changing world. Known globally as a maker of leading-edge technology, we enable our customers to easily transform their data into actionable information, shortening the lifecycle from the moment of change to action. Hexagon Geospatial provides the software products and platforms to a large variety of customers through direct sales, channel partners, and Hexagon businesses, including the underlying geospatial technology to drive Intergraph
® Security, Government & Infrastructure (SG&I)
industry solutions. Hexagon Geospatial is a division of Intergraph® Corporation. For more information, visit
www.hexagongeospatial.com.
Intergraph® Corporation is part of Hexagon (Nordic exchange: HEXA B). Hexagon is a leading global provider of
design, measurement and visualisation technologies that enable customers to design, measure and position objects, and process and present data.
Learn more at www.hexagon.com.
© 2014 Intergraph® Corporation. All rights reserved. Intergraph is part of Hexagon. Intergraph and the Intergraph logo are registered
trademarks of Intergraph Corporation or its subsidiaries. Hexagon and the Hexagon logo are registered trademarks of Hexagon AB or its subsidiaries. All other trademarks or servicemarks used herein are property of their respective owners. Intergraph believes the information in this publication is accurate as of its publication date. Such information is subject to change without notice. GEO – US – 0253A - ENG