Comparison of Windows IaaS Environments

Similar documents
Cloud IaaS Performance & Price-Performance

Cloud IaaS Performance & Price-Performance

Generational Performance Comparison: Microsoft Azure s A- Series and D-Series. A Buyer's Lens Report by Anne Qingyang Liu

Comparing major cloud-service providers: virtual processor performance. A Cloud Report by Danny Gee, and Kenny Li

Network Performance Between Geo-Isolated Data Centers. Testing Trans-Atlantic and Intra-European Network Performance between Cloud Service Providers

2016 TOP 10 CLOUD VENDOR BENCHMARK. EUROPE REPORT Price-Performance Analysis of the Top 10 Public IaaS Vendors

Cloud Analysis: Performance Benchmarks of Linux & Windows Environments

Performance Analysis: Benchmarking Public Clouds

Cloud Server Performance A Comparative Analysis of 5 Large Cloud IaaS Providers

IaaS Performance and Value Analysis A study of performance among 14 top public cloud infrastructure providers

Choosing Between Commodity and Enterprise Cloud

By Cloud Spectator July 2013

Cloud Vendor Benchmark Price & Performance Comparison Among 15 Top IaaS Providers Part 1: Pricing. April 2015 (UPDATED)

Part 1: Price Comparison Among The 10 Top Iaas Providers

SQL Server Consolidation Using Cisco Unified Computing System and Microsoft Hyper-V

The Secret World of Cloud IaaS Pricing: How to Compare Apples and Oranges Among Cloud Providers

DIABLO TECHNOLOGIES MEMORY CHANNEL STORAGE AND VMWARE VIRTUAL SAN : VDI ACCELERATION

The Secret World of Cloud IaaS Pricing in 2014: How to Compare Apples and Oranges Among Cloud Providers

Performance Analysis: Benchmarks of Bare-Metal & Virtual Clouds

Benchmarking Hadoop & HBase on Violin

HP ProLiant Gen8 vs Gen9 Server Blades on Data Warehouse Workloads

I/O PERFORMANCE COMPARISON OF VMWARE VCLOUD HYBRID SERVICE AND AMAZON WEB SERVICES

CLOUDSPECS PERFORMANCE REPORT LUNACLOUD, AMAZON EC2, RACKSPACE CLOUD AUTHOR: KENNY LI NOVEMBER 2012

MaxDeploy Ready. Hyper- Converged Virtualization Solution. With SanDisk Fusion iomemory products

The State of Cloud Storage

Oracle Database Scalability in VMware ESX VMware ESX 3.5

Accelerating Microsoft Exchange Servers with I/O Caching

Diablo and VMware TM powering SQL Server TM in Virtual SAN TM. A Diablo Technologies Whitepaper. May 2015

Virtuoso and Database Scalability

Rackspace Cloud Servers Analysis A comparison of cloud servers across Generations of Rackspace Cloud Offerings

Technical Paper. Moving SAS Applications from a Physical to a Virtual VMware Environment

WHITE PAPER Guide to 50% Faster VMs No Hardware Required

IBM Platform Computing Cloud Service Ready to use Platform LSF & Symphony clusters in the SoftLayer cloud

WHITE PAPER Guide to 50% Faster VMs No Hardware Required

8Gb Fibre Channel Adapter of Choice in Microsoft Hyper-V Environments

DOCLITE: DOCKER CONTAINER-BASED LIGHTWEIGHT BENCHMARKING ON THE CLOUD

Virtualization of the MS Exchange Server Environment

How To Store Data On An Ocora Nosql Database On A Flash Memory Device On A Microsoft Flash Memory 2 (Iomemory)

Performance Characteristics of VMFS and RDM VMware ESX Server 3.0.1

Performance Testing of a Cloud Service

Comparing Multi-Core Processors for Server Virtualization

BETTER PUBLIC CLOUD PERFORMANCE WITH SOFTLAYER

Virtualization Performance on SGI UV 2000 using Red Hat Enterprise Linux 6.3 KVM

The State of Cloud Storage

SQL Server Virtualization

JBoss Data Grid Performance Study Comparing Java HotSpot to Azul Zing

WHITE PAPER Optimizing Virtual Platform Disk Performance

Intel Cloud Builders Guide to Cloud Design and Deployment on Intel Platforms

What Is Microsoft Private Cloud Fast Track?

Amazon Web Services vs. Horizon

Cloud Computing Workload Benchmark Report

Amazon EC2 XenApp Scalability Analysis

EMC XtremSF: Delivering Next Generation Storage Performance for SQL Server

Deploying XenApp 7.5 on Microsoft Azure cloud

MS Exchange Server Acceleration

Microsoft Windows Server Hyper-V in a Flash

Design Considerations for Increasing VDI Performance and Scalability with Cisco Unified Computing System

Overview: X5 Generation Database Machines

Cloud Computing Performance. Benchmark Testing Report. Comparing ProfitBricks vs. Amazon EC2

Evaluation Report: Accelerating SQL Server Database Performance with the Lenovo Storage S3200 SAN Array

GRIDCENTRIC VMS TECHNOLOGY VDI PERFORMANCE STUDY

Cloud Storage. Parallels. Performance Benchmark Results. White Paper.

Scaling Database Performance in Azure

HP SN1000E 16 Gb Fibre Channel HBA Evaluation

Analysis of VDI Storage Performance During Bootstorm

Introduction. Need for ever-increasing storage scalability. Arista and Panasas provide a unique Cloud Storage solution

PARALLELS CLOUD SERVER

HP ProLiant BL660c Gen9 and Microsoft SQL Server 2014 technical brief

BENCHMARKING CLOUD DATABASES CASE STUDY on HBASE, HADOOP and CASSANDRA USING YCSB

Storage I/O Control: Proportional Allocation of Shared Storage Resources

Oracle Database Reliability, Performance and scalability on Intel Xeon platforms Mitch Shults, Intel Corporation October 2011

Colgate-Palmolive selects SAP HANA to improve the speed of business analytics with IBM and SAP

Maximizing SQL Server Virtualization Performance

Using Synology SSD Technology to Enhance System Performance Synology Inc.

The MAX5 Advantage: Clients Benefit running Microsoft SQL Server Data Warehouse (Workloads) on IBM BladeCenter HX5 with IBM MAX5.

Dell Virtualization Solution for Microsoft SQL Server 2012 using PowerEdge R820

HP Z Turbo Drive PCIe SSD

Removing Performance Bottlenecks in Databases with Red Hat Enterprise Linux and Violin Memory Flash Storage Arrays. Red Hat Performance Engineering

Cloud Computing Performance Benchmarking Report. Comparing ProfitBricks and Amazon EC2 using standard open source tools UnixBench, DBENCH and Iperf

Deep Dive: Maximizing EC2 & EBS Performance

UBUNTU DISK IO BENCHMARK TEST RESULTS

Maximum performance, minimal risk for data warehousing

Accelerating Server Storage Performance on Lenovo ThinkServer

Achieving a High Performance OLTP Database using SQL Server and Dell PowerEdge R720 with Internal PCIe SSD Storage

On- Prem MongoDB- as- a- Service Powered by the CumuLogic DBaaS Platform

Data Centers and Cloud Computing

Data Centers and Cloud Computing. Data Centers

2009 Oracle Corporation 1

Boost Database Performance with the Cisco UCS Storage Accelerator

OTM in the Cloud. Ryan Haney

Cloud Spectator Comparative Performance Report July 2014

Powered by Intel Cloud Technology

EMC VFCACHE ACCELERATES ORACLE

Performance Benchmark for Cloud Block Storage

Benchmarking Large Scale Cloud Computing in Asia Pacific

Transcription:

Comparison of Windows IaaS Environments Comparison of Amazon Web Services, Expedient, Microsoft, and Rackspace Public Clouds January 5, 215

TABLE OF CONTENTS Executive Summary 2 vcpu Performance Summary 2 Memory Performance Summary 2 Disk Performance Summary 2 Internal Network Performance Summary 3 Introduction 3 Why Does Performance Matter 3 Key Considerations 4 Test Results 5 vcpu Performance Results 5 Memory Performance Results 7 Disk Performance Results 9 Internal Network Performance Results 11 Conclusion 13 Further Study 13 About Cloud Spectator 14 Methodology 15 Test Setup 15 Test Selection 15 About the VMs 15 Testing Notes 16 1

EXECUTIVE SUMMARY Cloud Spectator was commissioned by Expedient to gauge the performance of virtual machines (VMs) on several public cloud providers. Amazon Web Services, Microsoft Azure and Rackspace were included in this study for comparison. Both performance and price-performance were examined to evaluate the value of each provider s VM. The purpose of the study was to understand the disparity of performance and value (defined as price-performance) between cloud providers for similarly sized VMs. The primary components of virtual servers, vcpu, memory, disk and internal network, were evaluated for each of the provider s VMs. vcpu PERFORMANCE SUMMARY For this study, Cloud Spectator evaluated vcpu performance by benchmarking the VMs using PassMark, a suite of benchmark tests for the Windows operating system, over the course of a 7-day test period. Pricing was examined in vcpu Performance Key Findings: Rackspace exhibited 16-27% higher vcpu performance Amazon s vcpu performance had the lowest variability Expedient offered the highest price-performance value for vcpu conjunction with the performance tests. MEMORY PERFORMANCE SUMMARY Memory was evaluated using the PassMark benchmark suite over a 7-day test period. Pricing was examined in conjunction with the performance tests. Memory Performance Key Findings: Amazon exhibited 16-48% higher memory performance Microsoft Azure s memory performance had the lowest variability Amazon offered the highest price-performance value for memory STORAGE PERFORMANCE SUMMARY Storage was evaluated using the PassMark benchmark suite over a 7-day test period. Persistent storage (offered as block storage ) was used in all storage tests. Pricing was examined in conjunction with the performance tests. Storage Performance Key Findings: Expedient exhibited.6x-32x higher storage performance Amazon s disk performance had the lowest variability Expedient offered the highest price-performance value for storage 2

Indexed Processor Score INTERNAL NETWORK PERFORMANCE SUMMARY Internal network performance was measured as the throughput between VMs within the internal environments of the cloud provider (measured using iperf). Throughput was monitored over the course of a 7-day test period. Pricing was examined in conjunction with the performance tests. Network Performance Key Findings: Expedient exhibited 39-15% higher internal network performance Amazon s internal network performance had the lowest variability Expedient offered the highest price-performance value for internal network INTRODUCTION 12 1 8 6 Processor Performance Across the IaaS Industry WHY DOES PERFORMANCE MATTER? Cloud infrastructure performance is a key consideration that should not be mistakenly overlooked; differences in performance outputs of virtual machines across the industry make it hard to compare across IaaS providers in a 4 2 standardized manner by simply examining features and/or pricing. Figure 1 on the left illustrates an example of the average processor performance from a sample of Cloud CSP 1 CSP 2 CSP 3 CSP 4 CSP 5 CSP 6 Service Providers (CSPs) as studied by Cloud Spectator. CSP Cloud Service Provder (CSP) 1 and CSP 6 (names removed) have a marked 3x difference in Figure 1: Variance in VM performance across the industry means that simply comparing features or pricing cannot make an apples-to-apples processor performance, giving CSP 1 a sizable advantage in comparison. many processor-intensive workloads. CSPs 2-5 exhibit a closer resemblance in processor performance, but also do not offer nearly as much processing power as CSP 1. Selecting the wrong provider to house an application can result in unnecessary overspending and/or application performance problems. Cloud Spectator examines infrastructure performance on both a hardware and an application layer. By examining the underlying hardware (in most cases, the virtualized hardware), Cloud Spectator can provide an understanding of the theoretical maximum and sustained performance of each component that comprises the server. Afterwards, running application tests helps determine which hardware bottlenecks occur first on the server for different types of applications. This proactively addresses performance concerns in a test environment before it can affect a production environment. Table 1 below lists the 3 hardware components studied in this project, and each purpose as a function in the server. 3

vcpu PERFORMANCE MEMORY PERFORMANCE STORAGE PERFORMANCE The performance of all applications is highly dependent on the vcpu. The vcpu is responsible for the processing and orchestration of all applications. While memory performance is not considered one of the key bottlenecks in performance for many applications, a subset of applications particularly HPC and in-memory databases is highly dependent on large sustained memory bandwidth. Because most applications and all data reside on the storage, having fast storage performance is a key consideration for smooth application performance in many cases. Table 1: A summary of each of the hardware components examined in this study and their functions within the server. PRICE-PERFORMANCE COMPARISONS Cloud Spectator s price-performance calculation, the CloudSpecs Score, provides information on how much performance is realized for each unit of cost. The CloudSpecs Score is an indexed, comparable score ranging from -1 indicative of value based on a combination of cost and performance. The value is scaled; e.g., a Cloud Service Provider (CSP) with a score of 1 gives 4x the value of a CSP with a score of 25. The CloudSpecs Scores in this report can only be compared with equivalent configurations; e.g., a 1vCPU VM on Provider A can only be compared to a 1vCPU VM on Provider B. The calculation of the CloudSpecs Score: 1. provider_value = [Provider Performance Score] / [Provider Cost] 2. best_provider_value = max{provider_values} 3. CSP s CloudSpecs Score = 1*provider_value / best_provider_value KEY CONSIDERATIONS When examining the results of these tests, please keep the following in mind: Testing was conducted on one specific VM size for each provider. Different VM configurations may yield different comparative results between the providers. Amazon, Microsoft and Rackspace offer fixed VM configurations with different resource focuses, while Expedient offers independently scalable resources. Prices used in the price-performance comparisons are up to date effective December 21, 214. Pricing may change for the specified VMs after the release of this report. The same VM was tested for each of the selected configurations on each of the providers. Users may experience different performance characteristics across different physical hosts. Factors such as user contention ( noisy neighbor ) or malfunctions of the physical hardware can cause less than optimal performance. The VMs selected were the base offerings across the providers; greater performance may be obtained on certain providers by paying for additional features/services. 4

CPU Score TEST RESULTS vcpu Performance Results The VMs selected for each provider contained four virtual cores or vcpus. The vcpus were tested using the PassMark suite of benchmark tests for Windows operating systems. The tests included mathematical operations such as integer and floating point calculations, as well as common CPU tasks such as compression and encryption of files. The aggregate results of the individual CPU tests were compounded together for an overall CPU Mark score. The performance, determined by the overall CPU Mark score, was then matched to the price of the VM to calculate the CloudSpecs Score for price-performance comparisons. CPU Performance 6 5 4 Amazon Expedient Microsoft Rackspace 3 2 1 Chart 1: The chart displays the vcpu performance of the provider VMs over the entire 7 day testing period. PROVIDER AVERAGE STDEV CV MAX MIN EXPEDIENT 4535 67 1% 4615 449 AMAZON 4145 16 % 4172 4124 MICROSOFT 4418 138 3% 457 3996 RACKSPACE 5262 162 3% 5352 4859 vcpu Performance (see chart above) According to the CPU Score from the PassMark test results, Rackspace achieved between 16% to 27% higher vcpu performance than Amazon, Expedient and Microsoft. Performance variability, as determined by the coefficient of variation (CV), was stable for all four providers who exhibited CVs of 3% or lower. 5

CloudSpecs Score The performance results were surprising because Expedient uses more advanced Intel Xeon E5-2695 v2 processors while both Amazon and Rackspace use the Intel Xeon E5-268 v2 processors. Rackspace s performance superiority compared to the same processor utilized by Amazon and the higher-level processors on Expedient was noteworthy. Microsoft uses the Intel Xeon E5-266 processors, which outperformed the newer processors utilized by Amazon as shown in the results. A possible rationale for the performance disparities given the aforementioned processor types was that contention for vcpu resources between users on each VM varied between provider to provider. CPU Price-Performance 1 9 8 7 6 5 4 3 2 1 1 69 68 6 Expedient Rackspace Amazon Microsoft Chart 2: The chart displays the vcpu price-performance of the provider VMs over the entire 7 day testing period. vcpu Price-Performance (see chart above) When considering the prices of the VMs along with performance scores in the calculated CloudSpecs Score, Expedient offered between 44% to 66% greater value than the three other providers included in this study. As evidenced by the graph above, Rackspace offered the second most value for vcpus, trailed by Amazon then Microsoft. Despite Rackspace s higher vcpu performance (see Chart 1), Expedient s lower price in combination with its relative performance offered slightly greater value than Rackspace for vcpus. 6

Memory Score Memory Performance Results The VMs selected for each provider offered a relative 2:1 ratio of 2GB memory to each 1vCPU. The memory was tested using the PassMark suite of benchmark tests. The tests included common memory tasks such as reading and writing from the RAM. The aggregate results of the individual memory tests were compounded together for an overall Memory Mark score. The performance, determined by the Memory Mark score, was then matched to the price of the VM to calculate the CloudSpecs Score for priceperformance comparisons. Memory Performance 25 2 15 Amazon Expedient Microsoft Rackspace 1 5 Chart 3: The chart displays the memory performance of the provider VMs over the entire 7 day testing period. PROVIDER AVERAGE STDEV CV MAX MIN EXPEDIENT 1283 51 4% 1377 1181 AMAZON 1895 167 9% 213 1498 MICROSOFT 163 26 2% 1675 1579 RACKSPACE 1572 18 7% 178 142 Memory Performance (see chart above) According to the Memory Score from the PassMark test results, Amazon achieved between 16% to 48% higher memory performance than Expedient, Microsoft and Rackspace. Performance variability of memory was relatively stable for Expedient and Microsoft with CVs of 4% and 2% respectively. Amazon and Rackspace were slightly higher with CVs of 9% and 7% respectively. A possible reason for the disparity in RAM performance between the providers could be due to the quantity of DIMM s populated in each physical server. Expedient populates 24 slots providing 384GB of RAM which changes the maximum frequency to 166MHz. 7

CloudSpecs Score Other service providers may only fill banks 1 and 2 for a maximum available amount of RAM of 256GB per server, changing the max frequency to 1866MHz. Memory Price-Performance 1 9 8 7 6 5 4 3 2 1 1 9 71 66 Amazon Expedient Microsoft Rackspace Chart 4: The chart displays the memory price-performance of the provider VMs over the entire 7 day testing period. Memory Price-Performance (see chart above) The analysis of the results revealed the price-performance values for memory, wherein Amazon offered between 11% to 51% greater value than the three other providers included in this study. By offering the highest performance (see Chart 3), despite charging a higher price compared to Expedient, Amazon was able to offer a higher overall price-performance ratio, which resulted in the highest CloudSpecs Score. Nonetheless, Expedient and Amazon offer similar levels of value for memory. Microsoft and Rackspace s memory performance were offset by higher costs, resulting in relatively low price-performance values. 8

Disk Score Storage Performance Results The storage performance for each provider was based upon testing of the persistent storage. Persistent storage can be included either as the primary storage or as on option in addition to ephemeral storage depending on the provider. Ephemeral storage puts the user at risk of losing data if the VM shuts off for any reason. Persistent storage maintains any stored data regardless of the life of any attached VMs. The storage was tested using the PassMark suite of benchmark tests. The tests measured the performance of sequential read, sequential write and random seek + R/W operations. The average results of the three storage tests were compounded together for an overall Disk Mark score. The performance, determined by the overall Disk Mark score, was then matched to the price of the VM to calculate the CloudSpecs Score for price-performance comparisons. Storage Performance 1 9 8 7 6 5 4 3 2 1 Amazon Expedient Microsoft Rackspace Chart 5: The chart displays the disk performance of the provider VMs over the entire 7 day testing period. PROVIDER AVERAGE STDEV CV MAX MIN EXPEDIENT 8897 559 6% 9479 756 AMAZON 916 27 3% 94 861 MICROSOFT 266 23 9% 297 28 RACKSPACE 5374 245 5% 5696 4927 Storage Performance (see chart above) The results of the storage tests show that Expedient achieved between 66% to 3239% higher storage performance than Amazon, Microsoft and Rackspace. Performance variability was stable for Amazon and Rackspace with CVs of 3% and 5% respectively. Microsoft s variability was slightly elevated with a CV of 9%. 9

CloudSpecs Score 1 9 8 7 6 5 4 3 2 1 Storage Price-Performance 1 36 8 2 Expedient Rackspace Amazon Microsoft Chart 6: The chart displays the disk price-performance of the provider VMs over the entire 7 day testing period. Storage Price-Performance (see chart above) As a result of testing the storage and examining prices of the VMs to find the price-performance values, the findings revealed that Expedient offered between 1.8x to 53x greater value than the three other providers included in this study. Rackspace offered the second highest value for storage, followed by Amazon then Microsoft. 1

Mbit/sec Internal Network Performance Results The performance of the internal network on each of the providers was tested by examining the throughput between VMs within the same data center. The network was tested using Iperf which sends data bi-directionally between two servers for a set amount of time. The performance, determined by the network throughput, was then matched to the price of the VM to calculate the CloudSpecs Score for price-performance comparisons. Internal Network Performance 18 16 14 12 1 Amazon Expedient Microsoft Rackspace 8 6 4 2 Chart 7: The chart displays the network performance of the provider VMs over the entire 7 day testing period. PROVIDER AVERAGE STDEV CV MAX MIN EXPEDIENT 1468 182 12% 1616 887 AMAZON 14 % 14 13 MICROSOFT 154 27 26% 1591 757 RACKSPACE 588 21 4% 67 531 Internal Network Performance (see chart above) The network throughput tests show that Expedient achieved between 39% to 15% higher network performance than Amazon, Microsoft and Rackspace. Performance variability of network was the most stable on Amazon with a CV of %. Expedient appears to be relatively as stable except for a drop in performance on December 23 rd, possibly due to increased network contention due to the Christmas holiday. Expedient s 12% CV value is largely due to the dip in performance. Microsoft displayed variable performance throughout the testing period with a CV of 26%. 11

CloudSpecs Score 1 9 8 7 6 5 4 3 2 1 Internal Network Price-Performance 1 51 44 24 Expedient Amazon Microsoft Rackspace Chart 8: The chart displays the network price-performance of the provider VMs over the entire 7 day testing period. Internal Network Price-Performance (see chart above) Taking into account the price of the VMs along with performance scores in the calculated CloudSpecs Score, Expedient offered between 1x to 3x greater value than the three other providers included in this study. As evidenced by the graph above, Amazon offered the second most value for network, trailed closely by Azure, offering 15% greater value than the latter provider. Although Microsoft performed 5% higher than Amazon (see Chart 7), the higher price offset the difference in performance and places Amazon slightly ahead for network price-performance value. 12

CONCLUSION The purpose of this study was to examine the differences in performance and value of the cloud offerings across various public cloud providers. Performance and price-performance of the servers were examined to determine the value offered, as determined by the price-performance ratios, albeit standardized into a general score for easier comparison (CloudSpecs Score). Performance must be examined first to determine the baseline amount of performance that will be offered by each provider. Pricing is then factored in to understand how much a user would need to be spent on each provider to obtain the given level of performance. The value of the cloud offerings is important for users to consider when determining the amount of resources needed for an application to run, and utilizing the optimal amount of spend for the selected resources. Consumers want to maximize their utility of the goods and services they purchase. The study revealed that Expedient offered the highest overall value for the majority of services that were examined. Expedient exhibited the highest performance in storage and network tests, while being second highest in vcpu and fourth in the memory tests. Amazon offered the either the best or second best value for vcpu, memory and internal network while Rackspace was second best for storage value. Variability in performance was low for all providers for vcpu, with CVs under 4%. Variability for memory and storage varied between providers, but CVs remained under 9%. Microsoft exhibited significantly high performance variability for internal network, with a CV of 26%. Network variability was elevated for Expedient, largely due to an outlier in performance during a small window within the testing period. Performance and value differences between providers were evidenced by this study. Providers such as Expedient have certain areas where they perform best, but also may offer the best value overall given the pricing of their services. At the same time, providers with more customizable offerings may allow you to purchase additional resources or performance services based on application needs. FURTHER STUDY Further studies may include more VM sizes and varying configurations across the providers. Some providers throttle performance based on the amount of resources allocated. As a result, performance may scale differently on certain providers based on different vcpu to memory ratios or size of the disks. Many providers allow you to pay for additional features and/or services that increase performance of the VMs. Users may be able to provision faster storage hardware, additional IOPS, larger network bandwidth, isolated physical environments, etc. 13

ABOUT CLOUD SPECTATOR Cloud Spectator is a cloud analyst agency focused on cloud Infrastructure-as-a-Service (IaaS) performance. The company actively monitors 2+ of the largest IaaS providers in the world, comparing VM performance (i.e., CPU, RAM, disk, internal network, and workloads) and pricing to achieve transparency in the cloud market. The company helps cloud providers understand their market position and helps businesses make intelligent decisions in selecting cloud providers and lowering total cost of ownership. The firm was founded in early 211 and is located in Boston, MA. 14

METHODOLOGY TESTING SETUP Cloud Spectator identified and provisioned equivalent sized VMs across each provider. The locations of the VMs were all in North America. The chosen benchmarking software was then installed on each VM and run twice a day for a period of 7 days. TESTS USED Test Category Description Nine separate CPU tests that are all aggregated into a final score. PassMark CPU Subtests include: Integer Math, Floating Point Math, Prime Numbers, Extended Instructions (SSE), Compression, Encryption, Physics, Sorting and Single Threaded. Seven separate memory tests that are all aggregated into a final score. PassMark Memory Subtests include: Database Operations, Read Cached, Read Uncached, Write, Available RAM, Latency, and Threaded. PassMark Disk Average of Sequential Read, Sequential Write and Random Seek + R/W tests. Iperf Network Transfers data bi-directionally between 2 nodes within 6 seconds, measured in megabits per second. PassMark test details and benchmark software can be found at: http://www.passmark.com/products/pt.htm ABOUT THE VMS All VM sizes were selected by Cloud Spectator, standardized as closely as possible by vcpus and memory. All VMs ran Windows Server 212 R2 operating systems. Two VMs of identical configurations were provisioned on each provider, one VM is used for the CPU, memory and disk tests, while the second VM is used in testing the internal network performance. SAMPLE SIZE EXPEDIENT AMAZON MICROSOFT RACKSPACE OFFERING NAME Level 15 Instance m3.xlarge D3 I/O1-15 vcpus 4 4 4 4 MEMORY 15GB 15GB 14GB 15GB BLOCK STORAGE 5GB 5GB 5GB 5GB US WEST DATACENTER Pittsburgh (N. EAST US DFW California) VM PRICE ($/HR) $.411 $.56 $.684 $.675* STORAGE ($/HR) $.14 $.6 $.2 $.34 *Rackspace charges a monthly Service Level fee on top of raw infrastructure costs. The lowest priced Service Level tier (Managed Infrastructure) was used in this study. 15

TESTING NOTES Testing was conducted for a period of 7 days. o The testing spanned from 12/18/14 to 12/24/14. o Tests were conducted twice per day for each day of testing; once in the morning and once in the afternoon. The average performance values from the entire testing period were used in the price performance calculations as the [Provider Performance Score] The performance variability was measured using the coefficient of variation (standard deviation / average) of values from the entire testing period. Persistent storage (also known as block storage) was tested instead of ephemeral storage (also known as local storage) due to the greater dependability of the data volume as well as the common user need to retain data regardless of the on/off state of the VM. 16