ºÝºÝߣ

ºÝºÝߣShare a Scribd company logo

Going the extra mile

?Download as PPT, PDF?
?
G
gojkoadzic

From the Gaming Scalability event, June 2009 in London (http://gamingscalability.org). Today's electronic markets are demanding lower and lower end-to-end latency. The acceptable time from an event occurring to reporting the effect of the event to consumers around the world has gone down to seconds and even sub-second in some cases. This presentation shows how the combination of an Enterprise Data Fabric inside the firewall and Push Technology for delivery over the Internet ("the last mile") can effectively address this problem and turn it into competitive advantage. Christian James is the product manager of Push Technology's high performance and low latent messaging product suite Diffusion. Christian leverages his knowledge and experience of the delivery of e-Commerce/betting and trading platforms, within the sports and financial markets, to expand and align Push Technology's product offering to the gaming fraternity. Prior to joining Push Technology Christian was head of IT for Cantor Spreadfair, a peer to peer spread betting exchange, and, prior to that, a lead within Deutsche Bank's FX e-Commerce team responsible for the design and development of Deutsche Bank's award winning autobahn?FX e-Commerce trading platform. Mike Stolz is vice president of architecture and strategy for financial services at GemStone Systems. In his role, Stolz leverages his expertise in targeting, developing and delivering innovative technology solutions to expand GemStone's global financial services offering and cultivate its growing capital markets division. During the ten years prior to joining GemStone Stolz served as director and chief architect of Merrill Lynch's global markets and investment banking debt division. In this role, Stolz was responsible for the design and development of trading systems and trading support systems for Fixed Income, Currencies, Commodities, Liquidity and Risk.

1 of 26
Going The Extra Mile Christian James Mike Stolz Product Manager VP of Architecture Push Technology Ltd GemStone Systems Inc.
Agenda ? Demands within the betting industry ? What is latency ? The cost of latency ? Combating latency ? GemFire/Diffusion solution ? Q&A
Demanding and competitive environment Technology drives competitive advantage
Demanding business requirements and drivers ? Speed, Stability, Scalability ? Increase the number of markets offered ? The ability to offer more volatile markets ? Deliver to more channels ? Improved marketing/advertising capabilities ? Minimize unmatched orders/bets ? Global distribution and real-time replication ? Localized Data creating better user experience ? Cost effective
Latency is everywhere What is latency?
Total latency equals Disk access + Sender overhead + Time of flight + Message size/Bandwidth ** + Receiver overhead ** Potentially misleading
E-Betting flow¡­ Bet rejected No No 100s of Web Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Servers Book Odds Position/ Risk Monitor Repository Database
¡­has many sources of latency Bet rejected No No 100s of Web Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Servers Book Odds Position/ Risk Monitor Repository Database
Why worry about it It Costs!
The cost of ignoring it is high, and its not just trading systems ? Amazon ¨C every 100ms of latency cost them 1% in sales ? Google ¨C an extra 0.5 seconds in search page generation time dropped traffic by 20% ? Financial ¨C ¡°If a broker¡¯s electronic trading platform is 5 ms behind the competition it could loose them at least 1% of their flow ¨C that¡¯s $4 million in revenues per ms¡± (TABB Group)
Latency reduces revenue, causes losses and results in a poor end-user experience Lost ??? Bet Rejected (Lost Revenue ???) Difference in odds X amt . No No 100s of Web Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Servers Book Odds Position/ Risk Monitor Repository Database
Money is lost as bets are rejected or taken at poor prices Odds generated, odds offered 10 to 1 Bet placed for ?10 @ 10 to 1 Bet booked, odds now being offered 5 to 1 End client wins Payout is ?100 versus ?50 In-running greatly amplifies this effect
Operating costs are high as architecture is inefficient Lost ??? Bet Rejected (Lost Revenue ???) Difference in odds X amt . No No 100s of Web Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Servers Book Odds Position/ Risk Monitor Repository Database
Budgets are being wasted on OPEX rather than advancing the business via CAPEX ? 60-70% of IT budgets spent on operational costs (CIO Magazine) ? 5% of IT budgets spend on energy (BBC) ? 1996-2006 no. of servers in London increased from 6 ? 28m. ? Avg. power consumption of each increased from 150 ? 400 watts. ? In 2007 it was predicted London would run out of power in 2009
¡°Latency Exists, Cope!¡± (Dan Pritchett) Combating it: Hardware or software?
Moore¡¯s law hasn¡¯t held up for hardware speed gains, they are slowing Speed of light Network Speed CPU Clock Speed, Moore¡¯s law Ram Speed (limited to distance from CPU) still exists (for how long) but for different reasons ¨C plus Memory/RAM Wall broken Desk I/O PS Date
Bandwidth should also be a cause for concern¡­ ? ¡°Net bombs¡± are gobbling up the internet ¨C YouTube = using the same amount of traffic as the entire internet in 2000 ¨C iPlayer 5% of UK internet usage. ? Monthly internet traffic is running at 8 exabytes (million trillion) ¨C one exabyte is equivalent of 50,000 years of DVDs ? 2007 traffic grew by 75%, capacity by 45% ? Brownouts/jitters are due to start in 2010 ? What is the alternative?
How we tackle it Faster software running on fewer machines "Hardware can give you a generic 20 percent improvement in performance, but there is only so far you can go with hardware. ¡° Rob Wallos, global head of market data Citi
GemFire & Diffusion kill inefficiencies reducing costs and increasing revenue Lost ??? Bet Rejected (Lost Revenue ???) Difference in odds X amt . No No Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Diffusion Book Position/ Risk Monitor Odds Repository GemFire Database GemFire looks after the first miles¡­ ¡­and Diffusion the last
In essence, a suite of low latency communication products ? GemFire allows you to create a highly resilient, elastic, enterprise data fabric to improve performance while simplifying your architecture. ? GemFire enables you to safe-store and receive data and events to and from any back-end data source and pass notifications off to Diffusion. GemFire ¨C Enterprise Data Fabric
In essence, a suite of low latency communication products ? Diffusion allows you to create a two-way secure real-time online channel between an organisation and its audience. ? Diffusion enables you to push and receive data and events to and from any current, or future, ¡°net¡± connected device including the web browser. Diffusion ¨C the power behind the Net
Smaller messages optimally delivered, reducing latency and foot print ? Best Message Delivery (BDM) ¨C Conflation ¨C High/Low Water Marks ¨C Throttling (could be used to level the ¡®playing field¡¯) ? Compression ? Your own message format avoids unnecessary transformation/serialization ? Only send deltas - reduces size/amount of data sent ? Hierarchical topics - reduces size/amount sent to finest grain ? Connecting various devices ¨C browser/mobile/interactive TV etc. ? These techniques can reduce bandwidth consumption by up-to 80% and help achieve a better end-user experience
Hierarchical topics, in context of a betting exchange
Strong focus on performance & efficiency, driven from the server ? NIO technology ensures data moves through Diffusion & GemFire as quickly as possible ? Zero Copy ? Zero Fan-out ? Bi-directional, no need to open another socket when sending information (executing a bet) back to the server as you would with a web request ? Scaling the application (massive number of clients on a single server) ? These approaches can massively reduce your hardware requirements
Global Edge Caching ensures most efficient client connectivity without replicating the whole application environment
Questions & Answers

More Related Content

Going the extra mile

  • 1. Going The Extra Mile Christian James Mike Stolz Product Manager VP of Architecture Push Technology Ltd GemStone Systems Inc.
  • 2. Agenda ? Demands within the betting industry ? What is latency ? The cost of latency ? Combating latency ? GemFire/Diffusion solution ? Q&A
  • 3. Demanding and competitive environment Technology drives competitive advantage
  • 4. Demanding business requirements and drivers ? Speed, Stability, Scalability ? Increase the number of markets offered ? The ability to offer more volatile markets ? Deliver to more channels ? Improved marketing/advertising capabilities ? Minimize unmatched orders/bets ? Global distribution and real-time replication ? Localized Data creating better user experience ? Cost effective
  • 6. Total latency equals Disk access + Sender overhead + Time of flight + Message size/Bandwidth ** + Receiver overhead ** Potentially misleading
  • 7. E-Betting flow¡­ Bet rejected No No 100s of Web Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Servers Book Odds Position/ Risk Monitor Repository Database
  • 8. ¡­has many sources of latency Bet rejected No No 100s of Web Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Servers Book Odds Position/ Risk Monitor Repository Database
  • 9. Why worry about it It Costs!
  • 10. The cost of ignoring it is high, and its not just trading systems ? Amazon ¨C every 100ms of latency cost them 1% in sales ? Google ¨C an extra 0.5 seconds in search page generation time dropped traffic by 20% ? Financial ¨C ¡°If a broker¡¯s electronic trading platform is 5 ms behind the competition it could loose them at least 1% of their flow ¨C that¡¯s $4 million in revenues per ms¡± (TABB Group)
  • 11. Latency reduces revenue, causes losses and results in a poor end-user experience Lost ??? Bet Rejected (Lost Revenue ???) Difference in odds X amt . No No 100s of Web Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Servers Book Odds Position/ Risk Monitor Repository Database
  • 12. Money is lost as bets are rejected or taken at poor prices Odds generated, odds offered 10 to 1 Bet placed for ?10 @ 10 to 1 Bet booked, odds now being offered 5 to 1 End client wins Payout is ?100 versus ?50 In-running greatly amplifies this effect
  • 13. Operating costs are high as architecture is inefficient Lost ??? Bet Rejected (Lost Revenue ???) Difference in odds X amt . No No 100s of Web Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Servers Book Odds Position/ Risk Monitor Repository Database
  • 14. Budgets are being wasted on OPEX rather than advancing the business via CAPEX ? 60-70% of IT budgets spent on operational costs (CIO Magazine) ? 5% of IT budgets spend on energy (BBC) ? 1996-2006 no. of servers in London increased from 6 ? 28m. ? Avg. power consumption of each increased from 150 ? 400 watts. ? In 2007 it was predicted London would run out of power in 2009
  • 15. ¡°Latency Exists, Cope!¡± (Dan Pritchett) Combating it: Hardware or software?
  • 16. Moore¡¯s law hasn¡¯t held up for hardware speed gains, they are slowing Speed of light Network Speed CPU Clock Speed, Moore¡¯s law Ram Speed (limited to distance from CPU) still exists (for how long) but for different reasons ¨C plus Memory/RAM Wall broken Desk I/O PS Date
  • 17. Bandwidth should also be a cause for concern¡­ ? ¡°Net bombs¡± are gobbling up the internet ¨C YouTube = using the same amount of traffic as the entire internet in 2000 ¨C iPlayer 5% of UK internet usage. ? Monthly internet traffic is running at 8 exabytes (million trillion) ¨C one exabyte is equivalent of 50,000 years of DVDs ? 2007 traffic grew by 75%, capacity by 45% ? Brownouts/jitters are due to start in 2010 ? What is the alternative?
  • 18. How we tackle it Faster software running on fewer machines "Hardware can give you a generic 20 percent improvement in performance, but there is only so far you can go with hardware. ¡° Rob Wallos, global head of market data Citi
  • 19. GemFire & Diffusion kill inefficiencies reducing costs and increasing revenue Lost ??? Bet Rejected (Lost Revenue ???) Difference in odds X amt . No No Accept & Yes Yes Can User Valid Bet? Place Bet Afford Bet? Diffusion Book Position/ Risk Monitor Odds Repository GemFire Database GemFire looks after the first miles¡­ ¡­and Diffusion the last
  • 20. In essence, a suite of low latency communication products ? GemFire allows you to create a highly resilient, elastic, enterprise data fabric to improve performance while simplifying your architecture. ? GemFire enables you to safe-store and receive data and events to and from any back-end data source and pass notifications off to Diffusion. GemFire ¨C Enterprise Data Fabric
  • 21. In essence, a suite of low latency communication products ? Diffusion allows you to create a two-way secure real-time online channel between an organisation and its audience. ? Diffusion enables you to push and receive data and events to and from any current, or future, ¡°net¡± connected device including the web browser. Diffusion ¨C the power behind the Net
  • 22. Smaller messages optimally delivered, reducing latency and foot print ? Best Message Delivery (BDM) ¨C Conflation ¨C High/Low Water Marks ¨C Throttling (could be used to level the ¡®playing field¡¯) ? Compression ? Your own message format avoids unnecessary transformation/serialization ? Only send deltas - reduces size/amount of data sent ? Hierarchical topics - reduces size/amount sent to finest grain ? Connecting various devices ¨C browser/mobile/interactive TV etc. ? These techniques can reduce bandwidth consumption by up-to 80% and help achieve a better end-user experience
  • 23. Hierarchical topics, in context of a betting exchange
  • 24. Strong focus on performance & efficiency, driven from the server ? NIO technology ensures data moves through Diffusion & GemFire as quickly as possible ? Zero Copy ? Zero Fan-out ? Bi-directional, no need to open another socket when sending information (executing a bet) back to the server as you would with a web request ? Scaling the application (massive number of clients on a single server) ? These approaches can massively reduce your hardware requirements
  • 25. Global Edge Caching ensures most efficient client connectivity without replicating the whole application environment