Upload
dangtram
View
220
Download
4
Embed Size (px)
Citation preview
Contents Business Problem NFR Challenges Solution Technology overview Performance Monitoring Approach Outcome Lessons Learned
OverviewCurrent Situation
Interest Rate SwapsManual ProcessesMultiple sourcesTrading Mechanism risk to dealers
OverviewCurrent Market Environment
Introducing change in a volatile environmentChange vs execution confidence
Credit Crunch hits high streetCredit Crunch hits high street
Volatile markets
Volatile markets…………
‘‘Market can findMarket can findsolutions to sub primesolutions to sub primedebt marketsdebt markets’’
NFR - Key ChallengesPerformance
Trade negotiation is time-sensitive, < 100 ms latency requiredYear one volume 20,000 msg/s scaling to 100,000 msg/s by year five.
ReliabilityNo loss of trade messages / transactions
Ease of IntegrationDealer & Distributors
Availability99.99%
Time to MarketGo Live in 9 Months
Technology - Pricing Example
Data Tier
Database
Middleware TierMiddleware Server Middleware Server
Gateway Tier
Quote Gateway Market Data GatewayTrade Gateway
Application Tier
TradeServices
AdminServices
PricingServices
ReportingServices
Dealers(INVESTMENT BANKS)
Distributors(REUTERS & BBG)
11
1. Dealer streaming in prices1. Dealer streaming in prices
222. Payload transformed and sent to2. Payload transformed and sent topricing servicespricing services
Market Data
(Composite Price)
3. Generates composite prices,3. Generates composite prices,twice every secondstwice every seconds
33
Technology - Products
Data Tier
Database
Middleware TierMiddleware Server Middleware Server
Gateway Tier
Quote Gateway Market Data Gateway
Application Tier
TradeServices
AdminServices
PricingServices
ReportingServices
Trade GatewayCameron US Cameron MDS
Fiorano JMS Fiorano JMS
Spring Spring Spring Spring
Oracle RAC
BEA WLRT BEA WLRT
BEA WLRT BEA WLRT
BEA WLRT BEA WLRT BEA WLRT BEA WLRT
Technology - Hardware
Trade & QuoteGateway
Market DataGateway
Middleware
Trade Services Pricing Services Admin Services Reporting Services
Op/Trade Databases
Tick/ReportingDatabases
HP DL360Dual Core 2xCPU 3GHz4GB RAM
HP DL580Dual Core 4xCPU 3.4GHz8GB RAM
KeyCameron GatewayCameron Gateway
WLRTWLRT
Fioarano Fioarano JMSJMSWLRTWLRT
SpringSpringWLRTWLRT
OracleOracleRACRAC
Technology - Product PerformanceCameron Universal & Market data ServerFix Servers
FIX is a standard communications protocol within the InvestmentBanking domain but is flexible enough to be customised
Cameron US/MDS can handle > 25,000 msg/s
Fiorano MQPure JMS ServerCan handle > 80,000 non-persistent msg/s
Technology - Product PerformanceWebLogic Real Time provided a huge performanceimprovement compared to standard Sun JVM
With – 15ms latencyWithout - significantly > 100ms latency
Underpinned key components in each tierFix Gateway ServersJMS ServersSpring
Made Java solution a feasible and good choice forthis systemSystem handling ~ 25,000 msg/s with 95th percentile latency~15ms
Performance Monitoring
Data Tier
Database
Middleware TierMiddleware Server Middleware Server
Gateway Tier
Quote Gateway Market Data Gateway
Application Tier
TradeServices
AdminServices
PricingServices
ReportingServices
Trade Gateway
JMX Monitoring Framework
MBean MBean
MBean MBean MBean
MBean
11
1. 1. MBean MBean Server for eachServer for eachcomponent/servicecomponent/service
2. Admin Services retrieves all2. Admin Services retrieves allstatistics periodicallystatistics periodically
22 MonitoringWeb App
3. Monitoring Web App displays3. Monitoring Web App displaysstatistics to usersstatistics to users
33
Performance MonitoringAdministration application provides detailedstatistics:
LatencyPer message typePer Component if required
ThroughputService/Server Status
Provides an easy to understand dashboard tothe LiquidityHub Staff.
Approach Reference Architecture Design completed in horizontal and vertical 4 architects, 4 ??
Responsibilities
Hands on ?
Expertise in the correct area
Outcome• Designed, built and delivered the system in the9 months
System meets year 1 peak performance targets Current latency figures 4 – 8ms
System can scale to much higher messagevolumes
Lessons Learned Create more testing tool
Distributor GUI test harness
Log search and view
Early prototyping helped de-risk the architecture
Consider virtualisation
Managing 17 organisations is not easy Weekly tech forums worked well
Dedicated relationship manager