FAST Protocol - FIX Adapted for Streaming

FAST Protocol Bandwidth and low latency Kevin Houstoun

Factors
Speed of light
299,792,458 m/s
299 m/us

People vs Computers
Computers nanoseconds People 140 - 200 milliseconds

Type of business
Solutions Smart Order Routing Proximity Hosting DMA Co location Care Orders Active Trading Human Involvement? Credit / Limit checking Booking Allocation Confirmation

670,616,629.4 miles per hour

1 ft / nanosec

Rapid Addition
FIX ~5 microseconds FAST ~ 1 microseconds

Measurement IP Stack
TCP/IP
Windows 30 microseconds Windows HPC 8 microseconds

Connectivity
FIX Hub FIX Direct IP Internet Private Switches, Routers and Hubs
2

UDP

Role of Market Data in Electronic Trading

The trend toward electronic trading is creating an explosion in market data volumes and peak message rates which present a challenge for even the best technology
Improved and ubiquitous trading technology is responsible for the rapid upward spiral in market data volume Broader use of standards have created a quick and easy way to connect to the market Market participants around the world make critical decisions on real-time information Without timely market data a trader is at an extreme disadvantage The pervasiveness of electronic trading is a tremendous stride forward but it also presents equally tremendous challenges

The market data catalyst

Market Drivers
Best executions Direct market access New Products New Participants Innovative Trading Strategies

Market Data
Market Structure Changes
Regulatory (Reg NMS, MiFID) Dynamic pools of liquidity

Innovation in Technology
Electronic Trading Algorithmic Models Reliance on Market Data for Decision Making Ubiquitous Trading Technology

Peak message rates outpacing effective solutions

Financial Information Forum Area Chart U.S. Equities Exchange Feeds - 1 Minute Peaks (Messages Per Second)
CTS
35,000

UTDF

CQS

NQDS

UQDF

ArcaBook

TotalView

OPRA Peak 1 Minute based on SIAC data (FIF Projections using Past Average % Change of 5.9%)
200,000

Source: Nasdaq, SIAC, Arca

30,000

175,000 150,000

25,000 Messages per Second

125,000
20,000

100,000
15,000

75,000 50,000 25,000 -

10,000

5,000

D ec Ja -01 n Fe -02 b M -02 ar A -02 pr M -0 ay 2 Ju -02 nJu 02 A l-0 ug 2 Se -0 p 2 O -02 c N t-0 ov 2 D -02 ec Ja -02 n Fe -03 b M -0 ar 3 Ap -03 M r-0 ay 3 Ju -03 nJu 03 Au l-0 g 3 S -0 ep 3 O -03 c N t-0 ov 3 D -0 ec 3 Ja -03 n Fe -04 b M -04 ar A -04 p M r-0 ay 4 Ju -04 nJu 04 A l-0 ug 4 Se -0 p- 4 O 04 c N t-0 ov 4 D -04 ec Ja -04 n Fe -05 b M -05 ar Ap -05 M r-0 ay 5 Ju -05 nJu 05 Au l-0 g 5 S -0 ep 5 -0 5

CME Bandwidth Utilization Trend

8500000 8000000 7500000 7000000 6500000 6000000 5500000 Bandwidth (mbits) 5000000 4500000 4000000 3500000 3000000 2500000 2000000 1500000 1000000 500000 0 12/29/2004 Moving Avg Avg Daily Average Daily Peak Moving Avg Peak

Business demands are outstripping communications services and cost-effective bandwidth solutions
More products, more listings, More price levels, more depth/data Increased demands for low latency

Proprietary solutions are increasingly costly to maintain The industry is ripe for a radical shift
2/17/2005 4/8/2005 5/28/2005 7/17/2005 9/5/2005 10/25/2005

1 Fe 99 b 9 Ap -00 r Ju -00 n Au -00 gO 00 c D t-00 ec Fe -00 b Ap -01 Ju r-01 nAu 01 g O -01 c D t-0 ec 1 Fe -01 b Ap -02 Ju r-02 nAu 02 g O -02 c D t-0 ec 2 Fe -02 b A -03 p Ju r-03 nAu 03 g O -03 c D t-0 ec 3 Fe -03 b Ap -04 r Ju -04 n Au -04 gO 04 c D t-04 ec Fe -04 b Ap -05 Ju r-05 n Au -05 g O -05 ct D -0 ec 5 Fe -05 b Ap -06 Ju r-06 n Au -06 g06

Challenge of Market Data Growth - OPRA

Source: Financial Information Forum, Market Data Capacity Committee

6

Challenge of Market Data Growth - US Equities

FIX Response to an Industry Problem

The Financial Services Industry faces seemingly ever growing market data bandwidth requirements, existing open standard are unable to cope with this volume and the industry faces developing another round of proprietary standards.
2004
Q4 Market Data Optimization Working Group (mdowg) formed

2005
Q1 Initial protocol proposals submitted Q2 Proof of Concept (POC) development Q3 POC testing performed Q4 POC results released

2006
Q1 FAST 1.0 specification released Q2 FAST 1.1 development work started Q3 FAST 1.1 specification development Q4 FAST 1.1 specification released

MDOWG Charter & Objectives

What are the challenges of engineering a new Market Data Solution? Develop a comprehensive solution that will mitigate exponentially increasing market data volume across all industry sectors Re-engineer Market Data so that it is no longer the weak link in the trading cycle Provide real-time, low-latency feeds that can be effectively delivered to the average subscriber Shrink the gap between trading technology and market data technology Plan for advances in bandwidth capacity but hedge your bets Create a common, flexible, straightforward solution that will reduce integration costs

What are key aspects of developing a global market data solution? Architecture
Scalable, fault-tolerant Architecture that minimizes latency and hops while providing flexibility

Interface
Standardized market data interface that supports all market sectors and business models

Transport
Efficient means of dissemination that reduces overhead while providing a lightweight session layer

Data Layer
Optimized data-transformation layer that radically reduces the size of a market data payload Typically 100 bytes becomes 20 bytes

Book Management
Intelligent book management practices that utilize thin-content market data to manage a book
10

Flexible, Scalable, Standards-based Architecture

11

FIX Adapted for Streaming (FAST)

Implicit tagging
Removes the need for the tag= portion of tag=value without imposing the constraints that fixed-length protocols do. Requires specifying message templates Retains flexibility and extensibility Retains repeating group optimisation from FIX.4.4

Field encoding
Reduces data volume by leveraging the data affinities between messages
If I dont send you a value for Symbol on this Exec Report, use the value from Symbol on the previous Exec Report I sent you If I dont send you a TimeInForce on this Order, use the default value of 0 (day) from the template

Serialization
Applies binary encoding to the data Uses continuation bit encoding Eliminates need for explicit separators Uses a presence map to indicate which fields are present, and which ones are not Special bytes in the header contain the template identifier, so that the other side can deserialize and decode appropriately Also considering adding a start-of-frame delimiter or message length to facilitate reading
12

MDOWG Accomplishments (to date)

Multi-asset class requirements review Book Management Recommended Practices Multicast Transport Recommended Practices FIX Extensions (Draft) for Market Data FAST Serialization and encoding specifications Initiated a FAST proof of concept (POC) Winter 2005 Published POC Results www.fixprotocol.org
13

Proof of Concept Framework

Data sets provided by POC sponsors (selected for high volume days where possible and type of data feed)
Top of Book OPRA Full Depth ARCA, NOREX Aggregate Depth CME

Measurements pre-defined and coded into test harness Preliminary testing to determine optimal field encoding Strict version control of the code base enforced Data sets were redundantly and reciprocally tested by two separate engineering teams
primary and secondary responsibility assignments primary results were compared to secondary results checked for uniformity; discrepancies researched test results correlated and re-run until identical results

Code base changes resulted in re-testing

14

POC Phase 1A Results Summary

15

FAST Protocol Performance Native Format ARCA

Bandwidth Utilization ARCA Feed vs FAST
3500 3000 2500

OPRA
Bandwidth Utilization OPRA Feed vs FAST
6000

5000

Kbits/second

2000 1500 1000 500 0 08:30 09:00 09:30 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Kbits/second

4000

3000

2000

1000

0 09:30 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

ARCA

FAST

T1

T1x2

OPRA

FAST

T1

T1x3

CME
Bandwidth Utilization CME Feed vs FAST All Channels
1600

NorEx
NOREX Bandwidth Utilization
50

1400

40
1200

Kbits/second

Kbits/second

1000

30

800

20

600

400

10
200

0 8:00

8:30

9:00

9:30

10:00

10:30

11:00

11:30

12:00

12:30

13:00

13:30

14:00

14:30

15:00

0 8:30

9:00

9:30

10:00

10:30

11:00

11:30

12:00

12:30

13:00

13:30

14:00

14:30

15:00

15:30

16:00

CME

FAST

T1

Norex

FAST

16

FAST ARCA Compression Ratio

17

FAST vs. Native ARCA Bandwidth Utilization

18

FIX Adapted for STreamingSM (FAST ProtocolSM)

FPLs Market Data Optimization Working Group developed and demonstrated a series of techniques to radically reduce message size

Very timely solution for real world volume and bandwidth challenges

Number of trades, Avg trade size, Speed

Quote volumes doubling year after year Algorithmic trading engines and DMA driving volume and need for speed Stat-arb and other trading strategies predicated on speed

The FAST Protocol is another example of global market participants working together to solve real world needs

A true solution, combine:

Standardized message definition, business model, and content in FIX FAST Protocol applied to reduce message size and serve high-volume needs

19

FAST 1.1 Overview

One single specification containing all the pieces Transfer Encoding
Support of non US ASCII Strings (byte vector) Presence map extension (now zero or one bit per field)

Field Encoding
Additional field operators

Templates
XML based notation to provide unambiguous template definitions.

FAST 1.0
Subset of FAST 1.1 High degree of compatibility
20

FAST 1.1 Specifics

Transfer Encoding
Support of non US ASCII Strings (byte vector) Presence map extension (now zero or one bit per field)

Field Encoding
Additional field operators

Templates
XML based notation to provide unambiguous template definitions.

FAST 1.0
Subset of FAST 1.1 High degree of compatibility

21

Business Case for FAST, SA faces bandwidth challenges

Bandwidth costs are reduced in proportion to the reduction in data volume. Heres an example:
A feed is projected to use 36 Mbps in the coming year Assume a 40 Mbps line is $6000/mo and a 10 Mbps line is $3000/mo If FAST is able to deliver 80% compression it now fits on a 9 Mbps line Savings is $36K annually times the number of customer connections

Hardware costs materially reduced

Lowers CPU utilization Reduced data storage requirements Server farms and routers dont need to increase

Flexible Architecture
New fields added with little effort Ability to plug-and-play any market data format

22

FAST Summary

FAST POC completed a year ago FAST scalability demonstrated to very high volumes Committed Implementations
CME OPRA Archipelago, NYSE ISE

Proof of Concept
LSE OMX

Hedge fund and high volume community using it.

23

Factors
Journey
Idea Order

People vs Computers
Computers nanoseconds People 140 - 200 milliseconds

Type of business
Solutions Smart Order Routing Proximity Hosting DMA Co location Care Orders Active Trading Human Involvement? Credit / Limit checking Booking Allocation Confirmation

Speed of light
299,792,458 m/s
299 m/us

670,616,629.4 miles per hour

1 ft / nanosec

Rapid Addition
FIX ~5 microseconds FAST ~ 1 microseconds

Measurement IP Stack
TCP/IP
Windows 30 microseconds Windows HPC 8 microseconds

Connectivity
FIX Hub FIX Direct IP Internet Private Switches, Routers and Hubs
24

UDP

Resources
FIX website: fixprotocol.org FIX website: fixprotocol.org/fast Technical / Specifications - fixprotocol.org/specifications
Spec docs, FIXML resources, FIXimate, Repository Technical Presentations / Documentation

FIX Implementation Guide - fixprotocol.org/implementation-guide FIXGlobal FPLs quarterly publication with free subscription fixglobal.com Algorithms www.fixprotocol.org/working_groups/algowg Market Data www.fixprotocol.org/working_groups/mdoptwg Remember
Visitors see something; Registered members see more; FIX Contributors see all.

Get Involved!!

25