Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
7 / 7 Accelerating FSI Applications with state-of-the-art Grid Framework - Real-time risk analysis and pricing in the front office
Laurence Liew, Platform Computing Singapore GridAsia 2008, September, 2008
Leader in HPC
2,000 Customers worldwide
Years of profitable growth
Employees in 15 offices
16 500
1 Leader in HPC
4,000,000 Managed CPUs
Financial Services Customers
Partner Enablement
LOB LOB LOB LOB
LOB
Phase 1 Phase 2 Phase 3 Phase 4
A A A A A
LOB
LOB LOB
A A A A
A A A A
A A A A
A A A A
A A A A
Four Phases of Grid Adoption
Silo “Internal Cloud” Enterprise Analytics Grid LOB Grid
Been there done that
When will DS figure this out?
How do I catch up to my competitors?
The Holy Grail
Improve performance & response time of one app
• Reduce exposure to risk – perform risk calculations in real-time, with greater accuracy
• Increase product profitability
Multiple Apps Sharing a Utility Grid
Dramatically lower Total Cost of Ownership
• Enable a utility service model through sharing of computational power across applications and lines of business
• 1% of increased utilization saves millions of dollars
LOB
App
Grid
Business Demands
• Beat the Competition – Financial institutions are creating more complex
products and increasing volumes
• Manage Risk – Address new regulatory requirements and be able to
respond in real-time to global volatility
• Be Cost Effective – IT decisions are scrutinized for the business value
they add, the potential ROI to the enterprise, and the Total Cost of Ownership
Normal Application Execution Symphony Execution
Elapsed Time
Symphony provides business advantage sub-millisecond latency > 90 MB/sec throughput per client <5 seconds to re-allocate 2,000 CPUs
Imagine the impact that instantaneous response and nearly unlimited compute power could have on your business critical application!
Grid Computing: Imagine the Possibilities
Successful Grid Deployments
Exotic Derivative
Model Calibration
Vanilla Flow
Basket Product
Risk Analysis
Capital Market
Asset Management Hedge
Funds
Option Houses
Exchanges
Front Office Trading Investment Mgmt Exchange Pricing
Middle Office Risk/End of day report Market Research Financial Accounting
Back Office Clearance & Settlement Batch Reporting Batch processing
Grid 2.0: the Front Office
• Enabling breakthrough business strategies: – Pricing – re-price & re-assess
rapidly in fluid markets – Modeling, back-testing &
optimization – intra-day / on demand
– Real-time risk – run risk algorithms in near real-time, assess exposures “tick-by-tick”
– Algorithmic trading – robo-trades / first mover advantage
– Statistical Arbitrage – reduce risk / maximize returns
Backtest Selection Execution
Overnight Daily
B1 S1 E1 Bn Sn En
Daily
Reech AIM – A Back-testing Example
• Historical Environment
• Intraday Grid
JPMorgan Chase: Grid in the Front Office
• Challenge: Expanding compute capacity while reducing costs – Dedicated SMP hardware – expensive, not
scalable, provisioned for peak – Major front and middle office trading and risk
systems that were expensive to maintain, difficult to scale
• Solution: – Built a grid infrastructure combining 7 trading
and risk systems
• Several million dollars of savings • A new credit trading application was built in just 10 weeks instead of the 5 months
• $0.56 /CPU hr
A Grid Example: Excel Modeling
• By far the most widely used modeling and simulation tool in financial services and insurance firms
• You don’t need to be a disciplined programmer to use Excel successfully
• A significant source of an institutions proprietary, competitive differentiation
Equates to a nightmare for developers and architects alike when the business needs to grow…
Why Grid-Enable Excel Applications?
• Improve performance and decrease response time by distributing complex compute tasks to the grid – Increase product profitability – Reduce exposure to risk
• Reduce costs by leveraging commodity servers
• Improve availability and security by “hardening” Excel user environments
Addressing Service Distribution with Grid
ODBC accessible data source holding market and portfolio data
VB/VBA scripts
Client API
CO
M C# Code
CLR
API
C# Code
CLR
API
C# Code
CLR
API
C# Code
CLR
API
SSM
SIM
SIM
SIM
SIM
SD
Using grid API’s, developers can easily access managed services in heterogeneous environments
Just How Much Work is Involved?
• … for “embarrassingly parallel” stochastic and risk modeling, very little indeed
Initialize and Connect
(2 lines of code)
Create Session (2 lines of code)
Send inputs (1-2 lines per input)
Retrieve outputs (2 lines of code)
Client Service
Implement Service Interface
{your calculation logic} Preparation
Distribution
Aggregation
I Now Have an HPC Application…
“For quants and actuaries, grid is addictive – once they’ve had a taste,
they can never get enough…”
IT Challenges
Service levels and required
resources
Time of Day 7am 8am 9am 10am 11am 12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm
Environment 1
Required Resources: CPU, Memory, and Overhead
Website
Environment 2
Environment 3
Development
Finance, Payroll Multiple
domains, departments
& technologies
Automate to Reduce Costs
Time of Day 7am 8am 9am 10am 11am 12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm
Required Resources: CPU, Memory, and Overhead
Adjusted Resources: VMs, CPU, Memory, and Overhead with Automation
$ savings
Service levels and required
resources
Multiple domains,
departments &
technologies
VM O
rche
stra
tor
Website
Development
Finance, Payroll
Making Cost Reduction a Reality
Making Cost Reduction a Reality Costs
(1 Year) Data Center Silo Grids Utility Grid
Environment (CPU’s) 4,000 2,500 Server Consolidation – resource sharing
CAPEX Hardware + Chassis $20,000,000 $12,500,000 Movement to more commodity hardware
OS support $400,000 $250,000 Reduced OS Subscriptions
Storage $86,400 $86,400
OPEX Data Center (power, cooling)
$1,822,080 $1,138,800 Utility Grid = true Green Computing
Data Center (rental) $600,000 $600,000 not going to reduce the size of the DC
System Administrator $2,500,000 $1,750,000 More CPU’s managed per Admin
Application Dev/Support team
$4,000,000 $4,000,000
Total $29,408,480 $20,325,200
CitiGroup – Utility Grid
Platform Symphony Platform EGO Platform LSF
Credit & Risk Applications
Platform EGO
Platform Symphony
Platform LSF
Real-time Applications
Long Running Applications
Anti-Fraud Anti-Laundering Operational Risk Enterprise Market Risk
Accounting, Actuarial Analysis Converts Pricing &
Hedging FX Derivatives Pricing & Hedging
CRM, Data Mining, Credit Scoring Credit Derivatives,
Pricing & Hedging Counterparty Credit Risk • Shared resource grid – 15+ intra
& end of day pricing & risk apps
• Utilization up from 20% to +80%
• Simplified Infrastructure management
• On-demand processing power at significantly reduced cost
Major International Bank
SAS Data Integration Studio SAS Grid Manager Microsoft Windows CCE
HQ
• Significant increases in processing capacity & reliability
• Avoided need to purchase additional computer systems
• Reduced key sub-process 3 days → 1
• Data load process 10 days → 8 (→ 2 planned)
• Plan to scale to 10 terabytes data in 1 year
Improving Service Level, Cutting Cost
Platform Symphony Platform EGO
• Twenty-one applications on corporate Compute BackBone
• Cost reduced to 56¢ / application hour
• Millions of dollars in savings
Platform Product Portfolio
Time Critical Services
Sequential & Parallel Batch
Virtual Machines
On-demand
Hosted Desktops
On-demand
3rd Party Applications (Business &
Specialty ISVs)
Heterogeneous Enterprise Resources (Physical and Virtual)
Enterprise Applications
PLATFORM ACCELERATE PLATFORM MANAGE
Platform Symphony
Platform Analytics
Platform RTM
Platform Manager
Platform VMO
Platform LSF
GemFire for Symphony
Platform LSF MultiCluster
Platform Process Manager
Platform LSF License Scheduler
Platform EGO
Platform Symphony
• Developers have free access to the Symphony Developer Edition with no node restrictions and no time limits
• Developers can interact with their peers and share their expertise at the community site www.hpccommunity.org
Market Data Portfolio Data Yield Curve Data Historical Data etc..
Data Service
Compute Grid
Analyst running monte-carlo simulation with 100,000 market scenarios, each task computation requiring access to a large data set
The Data Access Bottleneck
• Compute nodes calling a single data service or database quickly overwhelm the database – Results in data starvation, and
utilization and performance drop dramatically
GEMFIRE ENTERPRISE
Compute grid running with a single coherent in-memory data cache
Database load(), store()
get(), put()
Platform Data Cache Solutions
• Enable compute nodes to seamlessly access data locally through a scalable, reliable, high-performance in-memory cache
Platform Computing Unique Benefits
• Speed – The lowest-latency HPC service-oriented middleware – 5 x 10 times faster than alternative environments
• Openness – Easy for developers to build and deploy their HPC service-
oriented applications – Download, install and run an application in under an hour – Free to download with no time or node-count restrictions
• Cost Savings – Cut your silo server farms in half – The only true utility computing model
Questions ?
Contact email: [email protected] www.hpccommunity.org www.platform.com