Vern Brownell CTO

1
Building IT Infrastructure for the Next Bull
Market Scalability, Performance and Resilience

• Vern Brownell
  CTO Founder
  Egenera, Inc.

2
Agenda

• Introduction and infrastructure context
• Generations of systems infrastructure management
• The myth of High Availability and 5-nines
• Customer View
• Capital market trends and technology implications
• High Availability and Fault Tolerance
• A need for data center re-engineering and a Linux
  strategy
• The real world challenge and the solution

3
Traditional InfrastructureLengthy deployment,
complexity, multiple failure points
Server 1
Server 2
Server 3
SystemManagement
Clustering
Network
Storage
AC Power
4
Generations of Distributed Computing Management

• Scripts
• Network Management tools
• Systems Management software suites
• Provisioning and Deployment tools
• Holistic Provisioning tools
• Data Center Virtualization

5
The Myth of 5-Nines AvailabilityTypical
Infrastructure Implementation
INTERNET
Load Balancer
KVM Switch
Firewall
2-Way UNIX Servers(Web Server)
Ethernet Switch
4-Way UNIX Servers(Application Server)
Fibre Channel Switch
Firewall
4-Way UNIX Servers(Database)
Power
6
Todays Options

• Standardized configurations and tools base
• Look at emerging deployment software products
• Movement towards horizontal scaling
• Linux and Intel
• Egeneras BladeFrame and Processing Area Network
  Architecture

7
(No Transcript)
8
Capital Market Trends and Technology Implications

• Globalization
• 24 X 7 accessibility
• provide global monitoring of business critical
  clients
• Technology consistency
• lead to uniformity of tools and
  reliability/scalability on global basis
• Increased sophistication of clients
• Clients want ability to price the entire universe
  of options
• they want the pricing done in real-time,
  increasing the needs in compute processing and
  availability

9
Capital Market Trends and Technology Implications

• Despite economy slowdown, order transaction
  volumes are consistently rising
• Transactions up from 3M/day to 50M/day in less
  than a year
• Higher volumes are attributed to
• increase in derivatives and proprietary business
• shift from a spread based model to commission
  based
• Need much more resilient and scalable
  architecture for faster response to business
  needs
• Downtime of a firms trading system - millions
  can be lost in seconds
• this necessitates high availability systems and
  will drive rethinking of existing system
  architectures

10
Fault Tolerance High Availability

• Superior High Availability solutions were
  available, but werent widely used as decisions
  were made on market share vendor relationships
• Fire fighting is costing too much money
• Technology decisions will shift to being based on
  High Availability due the increasing volumes, and
  the need to operational 24x7
• Companies have learned the hard way they cant
  afford to have down time
• Need for self-healing clustering to minimize
  hardware failures

11
A Need for Data Center Re-EngineeringVirtualizati
on

• Improve cost, space/thermal effectiveness
• Rapid Deployment
• Simplified Management
• Built-in selectable High Availability
• High Reliability
• Lower cost of ownership
• Eliminate unnecessary components and
  the associated cost
• Pools of processing elements interconnected
  with pools of storage

12
Real World ApplicationProgram Trading Application

• Application
• Program Trading Application Processing
  Ticks/Second
• Technical Details
• The application was a multi-threaded application
  with 5 or less threads. The basic workload of the
  program was to process as many ticks/second as
  possible and run a pricing model on each tick.
  The application was highly compute bound as the
  workload increased causing a loss of revenue to
  the company.
• Critical Issue
• Desire to increase the throughput of the
  application but was faced with significant costs
  to upgrade the hardware to the latest Sun
  hardware.
• Reason
• The application was written to run under
  proprietary Solaris 2.6. Locked into this
  platform, few alternatives existed to improve
  performance of the application without a
  significant capital expenditure.

13
Real World ApplicationProgram Trading Application

• Vision
• Determine how to port the application to LINUX
  but maintain the RAS of present UNIX solution.
• Solution
• The code was ported to Linux and run on a single
  4-way Egenera Processing Blade. The application
  would fail-over to the global blade pool.
• Blade Frame scales better to distributed mode of
  application as 23 distinct servers can be
  configured on BladeFrame leaving one for high
  availability
• With Egenera solution, the TCO for horizontally
  scaling this application is much less than a
  complex, SMP or network based solution from SUN
• Results
• Capital cost reduction of 70 and 105 increase
  in calcs/sec

14
QA