As interest in cloud solutions and their use with enterprise applications has increased, MavenWire has taken a lead in implementing and benchmarking several instances of OTM using Amazon Web Services (AWS) and Elastic Cloud Compute (EC2). This presentation outlines how the instances were set up and configured; potential benefits of OTM in the cloud; cost and performance comparisons between the cloud and "traditional" server configurations; areas of concern and issues to be aware of when implementing OTM in the cloud. In addition, we will also outline what we believe the future direction of cloud OTM will be, as well as where we believe it is best suited to customer needs.
2. The Cloud
The Cloud is a set of services and
technologies that delivers real-time and on-
demand computing resources
Software as a Service (SaaS) delivers pre-
configured applications, usually through web
browsers
Platform as a Service (PaaS) delivers a
solution stack (like LAMP) tailored to certain
application types
Infrastructure as a Service (IaaS) delivers
complete server and network infrastructure
on-demand hosted by a cloud provider
3. Cloud Providers
Amazon AWS
Most popular and largest provider
PaaS and IaaS solutions
Large number of Cloud datacenters and
services
Rackspace
Offers Windows and Linux Cloud servers
Hybrid cloud model allows for half cloud,
half physical infrastructures
Microsoft
Windows Azure – Runs Windows and
Linux
Has IaaS and PaaS offerings
Many other providers, including leading
commodity hardware manufacturers
4. Benefits to the Cloud
Costs
No capital expenses, pay as you go
Scale on demand
Ease of maintenance, simplified infrastructure
Agility in responding to business needs
Instances dedicated to UAT, new projects,
patch/upgrade testing created on demand
Scripted deployments for fast server
creation and application installation
New projects can have server assets in
place in hours versus weeks or months
5. Downsides to the Cloud
Downtime Risks
Amazon AWS major outages
April 2011 - 36 hours, US East
August 2011 – 1 hour, US East
June 2012 – 6 hours, 14 hours, US
East
Audit and regulatory requirements
Major cloud providers have SSAE16/SAS70
reports, and are PCI-DSS Level 1 certified
Application expertise
High throughput, high performance cloud
offerings are not as fast as traditional
hardware
Cloud server configurations are limited
13. Costs - Yearly On Demand
$70,000.00
$60,000.00
$50,000.00
$40,000.00
Year 1
Year 2
$30,000.00 Year 3
$20,000.00
$10,000.00
$0.00
Mid-tier hardware Upper-tier hardware Amazon AWS Large Amazon AWS High IO Rackspace Large
Ongoing costs for hardware include power, colocation, and bandwidth
2 application/web and 1 database
configuration
14. Costs – AWS Reserved
$70,000.00
$60,000.00
$50,000.00
$40,000.00
Year 1
Year 2
$30,000.00 Year 3
$20,000.00
$10,000.00
$0.00
Mid-tier hardware Upper-tier hardware Amazon AWS Large Amazon AWS High IO Rackspace Large
Ongoing costs for hardware include power, colocation, and bandwidth
2 application/web and 1 database
configuration
15. Costs – 3 Year TCO On Demand
$140,000.00
$120,000.00
$100,000.00
$80,000.00 Year 3
Year 2
$60,000.00 Year 1
$40,000.00
$20,000.00
$0.00
Mid-tier hardware Upper-tier Amazon AWS Amazon AWS High Rackspace Large
hardware Large IO
Ongoing costs for hardware include power, colocation, and bandwidth
2 application/web and 1 database
configuration
16. Costs – 3 Year TCO with AWS Reserved
$120,000.00
$100,000.00
$80,000.00
Year 3
$60,000.00 Year 2
Year 1
$40,000.00
$20,000.00
$0.00
Mid-tier hardware Upper-tier hardware Amazon AWS Large Amazon AWS High Rackspace Large
IO
Ongoing costs for hardware include power, colocation, and bandwidth
2 application/web and 1 database
configuration
17. Benchmarks
DaCapo – Simulates single threaded loads
similar to bulk plans
VolanoMark – Simulates multi-threaded, high
subsystem I/O loads similar to agent
processing, also simulates web traffic
HammerOra – TPCC style Oracle OLTP
database benchmark, 70% read 30% write
18. DaCapo
25000
20000
Time (Milliseconds)
15000
Average Score
10000
5000
0
Rackspace Large Amazon AWS Large Amazon AWS High IO Mid-tier hardware Upper-tier hardware
Lower score is better
19. VolanoMark
450000
400000
350000
Connections per second
300000
250000
Average Score
200000 Average Per Core
150000
100000
50000
0
Rackspace Large Amazon AWS Large Amazon AWS High IO Mid-tier hardware Upper-tier hardware
Higher score is better
20. HammerOra
350000
300000
250000
Transactions per Minute (Higher is better)
200000 Amazon AWS Large
Amazon AWS High IO
Rackspace Large
150000 Mid-tier hardware
Upper-tier hardware
100000
50000
0
1 2 4 8 12 16 20 24 28 30 32 34 36
Virtual Users
21. Cost vs Performance - DaCapo
30
25
20
15 1 Year On Demand
1 Year Reserved
10
5
0
Rackspace Large Amazon AWS Large Amazon AWS High IO Mid-tier hardware Upper-tier hardware
Higher score is better
22. Cost vs Performance - VolanoMark
70
60
50
40
On Demand
AWS Reserved
30
20
10
0
Rackspace Large Amazon AWS Large Amazon AWS High IO Mid-tier hardware Upper-tier hardware
Higher score is better
23. Cost vs Performance - HammerOra
14
12
10
8
On Demand
AWS Reserved
6
4
2
0
Rackspace Large Amazon AWS Large Amazon AWS High IO Mid-tier hardware Upper-tier hardware
Higher score is better
24. Overall Cost vs Performance
AWS reserved instances make current Cloud
cost/performance exceed hardware in some
cases
Database performance per dollar spent is
higher on hardware
Storage I/O is the leading factor
Cost vs performance plays to the Cloud’s
existing strengths – horizontally scaled
applications
RDBMS and other applications that benefit
from vertical scale are less cost efficient
currently in the cloud
25. Disaster Recovery on AWS
Internet
Active DNS
Elastic Load Balancing Elastic Load Balancing
Route 53 Hosted Zone
EC2 Instance EC2 Instance EC2 Instance EC2 Instance
OTM App/Web 01 OTM App/Web 02 OTM App/Web 01 OTM App/Web 02
Security Group Security Group
EC2 Instance EC2 Instance
Database 01 Database 01
Mirroring / Replication
EBS Volume EBS Volume
Security Group Security Group
Availability Zone Availability Zone
US East 1a US West 1a
27. OTM Benefits and Usage
Development and Test Systems – Agility of
the cloud without the need for high
performance
Lower costs if reserved AWS servers are
Upgradeand if servers are powered off off-
used testing – Test new OTM versions
hours
without impacting existing development cycles
Disaster Recovery
Running versus non-running billing for
AWS
DR system is potentially lower throughput
Replicate databases, do not launch app
Training – Trainneeded cloud systems to
servers until users on
avoid impacting development cycles
28. OTM Benefits and Usage Con’t
Vendor certification/POC – Validate new
OTM related products with lower startup costs
High Performance Production – Cloud
performance still lags behind hardware
Support – Cloud technology is still new, bugs
and support difficulties may exist
Amazon and Oracle joint support
agreement for EC2 applications
Amazon RDS and Oracle
Future licensing
Troubleshooting Opacity – Opacity to
upstream issues can make troubleshooting
OTM performance more difficult
29. Future Cloud Growth
AWS prices are reduced 2-3 times per year,
on average
Amazon High I/O instance is the benchmark
for near-future Cloud performance
Google has joined the Cloud market with
Google Cloud Platform
Growth trends through 2010 show a faster
decrease in Cloud resource prices than
corresponding hardware, excluding storage
Future generations of enterprise applications
will be tailored to cloud deployments, both
public and private