Latency In the Real World The Game Of Truth Or Dare

Latency In the Real World
The Game Of Truth Or Dare
Donal Byrne | CEO Corvil |
Sep 29th, 2011
Corvil Experience In Low-Latency Trading
Buy Side
•
Buy-Side
Over 20 market maker and
proprietary trading firms
Other
•
9 of the G14 group of broker/dealers
monitor order flow performance
•
7 out of the top 10 investment banks
as active customers*
•
18 additional tier 2 banks and brokers
•
14 exchanges including 10 major
global exchanges
*Source FT.com League Tables, % of Equities Fees collected in Q1 2011
Sell-Side
Providers
Trading
Network
Exchanges
The Latency “Truth Or Dare” Game
• The order response time for a leading (low-latency) exchange in the US was
measured over the trading day using Corvil at the member / exchange
demarcation
• The following statements were both simultaneously TRUE as it relates to the
description of latency performance during the trading day:
• FROM EXCHANGE PERSPECTIVE: “Today, we were able to execute your orders
achieving 166 microseconds of latency.”
• FROM MEMBER PERSPECTIVE: “Today, I saw some of my orders delayed up to
2.2 seconds, what happened?”
• LESSON: Our industry needs a common agreed method for specifying and
describing the latency performance of high speed trading services
• Corvil is working with the trading community and FPL standards body to drive
such a common specification
Better Latency Transparency
• An improved way to describe the trading day’s latency performance by the
exchange to the member might be
• BETTER – “Today our best case latency was 166 microseconds, worst case
latency was 2.2 seconds, and the average latency for all your orders was 323
microseconds”
• BEST – “Today, 99% of your orders during the busy 1 second were executed
within 1.89 milliseconds”
• When looking at latency performance claims it is critical to understand:
• Measurement granularity, and accuracy
• Measurement quantile – min, max, avg, %tile
• Measurement interval – trading month, week, day, busy 1 sec
• Measurement jurisdiction – end-to-end, within 4-walls, across component
CASE STUDY : SGX Door to Door Latency
CORVIL
CNE-5400
Matching
Engine
Order Entry
Secure Access GWY
Start #2
Start #1
Order
Processed
by
Matching
Engine
Co-Location
Rack/Servers
Order Ack
Stop #1 71µsec
Latency Round-Trip
Door-to-Door #1
Stop #2 78µsec
Rack-to-Rack #2
Proprietary API
Median
71µsec
78µsec
90th Percentile
79µsec
86µsec
Median
110µsec
117µsec
90th Percentile
207µsec
217µsec
FIX 5.0
Ongoing Focus Area For SGX
•
Remove jitter to provide the best predictable response time possible
3rd Annual Trading Architecture Asia2011 – Bob Caisley CIO SGX Presentation
CASE STUDY : EUREX The Key Measurements
Member Server
T1
ETS Gateway
T1w
Eurex Matching
Engine
T5
T3
T7
T9
T9wB
T2
T2w
T4
T6
T9wA
Corvil
Engine
A - Feed
EBS Publisher
B - Feed
T8
MEMBER MEASURED
EXCHANGE MEASURED
1.Tick To Order (Wire) = T1w - T9w(A or B)
1. Exchange Order to Order Ack = T4 – T3
2.Order To Tick (Wire) = T9w(A or B) – T1w
2. ETS GW Response Time = (T4 – T3) – (T6 – T5)
3.Order To Order Ack (Wire) = T2w – T1w
3. Eurex Matching Response Time = T6 – T5
4.Relative A vs B (Wire) = T9wA – T9wB
4. Order Miss Lag Time = T5 (Member Order
Time) – T7 (Aggressor Fill Time)
Closing Thought
Information About Speed Will Become More Valuable
Than Speed Itself.
THANK YOU
Questions & Discussion
WWW.CORVIL.COM