A Call to Arms – Seizing Control of Enterprise Grids Through Open Source and StandardsCondor Week, 2007
Mike RyanChief Technologist, Compute [email protected]
2
JPMorganChase Is One of the Largest Financial Institutions
RetailFinancialServices
InvestmentBank
Treasury& Securities
Services
Asset Management
InvestmentManagement
& PrivateBanking
Clients
Core Technology
Over 160,000 employees
3
JP Morgan’s Investment Bank is an industry leader offering thousands of clients expert advice and execution in all aspects of capital formation, growth and preservation.
The Investment Bank
People: 18,000 in IB; 4,000 in IBTechLocations: Offices in 50 countriesClients: More than 90% of Fortune 1,000Revenues: $12 billion
Who We Are Lines of Business
Global Syndicated LoansInterest Rate DerivativesAsset-Backed SecuritiesInvestment Grade Corporate DebtEmerging Markets International BondsGlobal Announced M&A
Leadership Positions
The Investment Bank
Fixed IncomeCredit & Rates MarketsExotics & HybridsInstitutional Equities, EDG, F&OGlobal M&AProprietary PositioningGlobal Commodities & CurrenciesGlobal Credit Risk Management
4
The Investment Bank – where we fit in
Lines of Business
Fixed IncomeCredit & Rates MarketsExotics & HybridsInstitutional Equities, EDG, F&OGlobal M&AProprietary PositioningGlobal Commodities & CurrenciesGlobal Credit Risk Management
Core IB Technology
•Architecture
• Solution Selection governance
• Engineering governance
• Best practices / patterns
• Technology Portfolio management
• Foundational Components
• Identity / Access Management
• Enterprise BPM
• Enterprise Message Bus
• Grid Computing
• AMQP
5
Investment banks have computationally intense problems at enormous scale – such as risk management and derivatives pricing – that must be completed in a relatively short period of time
Risk Management Investment banking is one of the most highly regulated industries, with formalized risk-
based capital requirements for Value at Risk (VaR) and portfolio credit risk.
These risks are modeled through stochastic processes such as binomial trees and monte-carlo simulations, which scale linearly with increased computational resources
A high degree of accuracy in these risk models can seriously impact the bottom line: Lowers the level of regulatory capital required
Increases understanding of sensitivity to market changes or credit events
Derivatives Pricing New derivative products are created with increasing frequency
The ability to accurately price them is critical to future success
Has to be fast (literally millions / hour)
Portfolio optimization for algorithmic trading
Why Grid?
6
The JP Morgan grid is known as the “Compute BackBone” (CBB), and consists of the application of over 12,000 shared, networked computers distributed globally.
The CBB’s Infrastructure Investment In 2004 Achieved A 60% To 70+% Utilization Versus The 15% Utilization Typically Observed In Line Of Business Dedicated Computing. (Business Week, October-18-2004).
The CBB’s Reliability Has Been Clearly Demonstrated With Only “Minutes” Of Downtime Since Inception In June/2003. (Over Five 9’s.)
The CBB was named Waters magazine’s IT Project of the Year (December / 2003), and Computerworld’s Best in Class for 2004.
JP Morgan’s Answer: the CBB
7
The Compute Grid
Spare Computers
Smart Software Layer
Rates Equities Credit Global Credit Risk Proprietary Positioning
How does it work?
Computing power is harnessed into a central utility – the grid
Grid management software coordinates the supply of computing power with demands of the business
Jobs are submitted by applications, then executed on the grid, with the results returned to the applications
8
How does it work?
Rates Equities Credit Global Credit Risk Proprietary Positioning
The Compute Grid
Smart Software Layer
Spare Computers
When one line of business needs more resources than usual, additional power can be drawn from under utilized machines
Compute Backbone can also add new computing capacity in 24 hours Our target for On Demand initiative is 1 hour
9
CBB Challenges
Issue For 2006? Issue For 2007?Topics
YESMore Flexible
Resource Utilization
YES, but Condor is being used in development environments to
increase utilization
Deterministic &
Opportunistic
Grid Scheduling
YES NO – solved with Condor
On Demand Grids YESYES, but progress is being made
… and Condor will be a factor
64 bit support YES NO – solved with Condor
10
CBB Challenges
Issue For 2006? Issue For 2007?Topics
YESImproved Data
CachingNO – bolted a cache onto our grid
Orchestration
(parent / child)YES NO – built it ourselves
Coordinated view
beyond single
compute cluster
YES NO – built it ourselves
11
Phase I was Condor on the Desktop First application is a stateless pricing calculator running on over two thousand
desktop machines for our Exotics & Hybrids business in London
Runs hundreds of thousands of pricing jobs per day
Currently in production
Phase II was Condor in the Datacenter First application is risk calculator for CDO products
Currently in production on 64 bit RHEL 4 in NY
Built a queuing layer on top of Condor to warm start pricers and eliminate 98% of Condor’s scheduling overhead
Built management tools to simplify grid operations
What have we done with Condor?
12
Vastly expand the presence of Condor across the bank
Manage resource allocation between applications in our Condor grid
Manage multiple Condor pools efficiently
Drive migration from legacy grid infrastructure to next generation platform
What are we doing with Condor next?
13
We need developer / customer driven ownership of standards definitions There is appetite amongst enterprise customers to drive them
Current standards focus on “job submission” Collecting resources together for a job Managing individual long-running parallel jobs
Need a broader scope Grid services are composed of code, processes, requests, data storage, queries,
events – all with different lifecycles The standards need to manage all those aspects Platform independence & virtualization More Security Features
Proposed standards concepts: manage services and inter-relations between services manage data submission / staging REAL TIME SCHEDULING STANDARDS!!!!
What are we looking for from the community?
Standards, standards, standards!!!!
14
Thank you!
Thanks to Miron, Todd, and the entire Condor Team
We’re always looking for bright, talented people
We’re also looking to partner with
Condor thought leaders
Please feel free to talk with any of us:
Mike Ryan Evan Junek
[email protected] [email protected]
(212) 622-4191 (212) 622-4559
Madji Lo
(212) 622-0651