Solace Systems The Evolution of Messaging The Rise of the Appliance

Solace Systems The Evolution of Messaging The Rise of the Appliance
Clive Andrews
Mat Hobbis
Obninsk, 2 March, 2013
LSE The focus beyond Low Latency
EXTENT Trading Technology Trends & Quality Assurance

Transcript

1. The Evolution of Messaging The Rise of the Appliance Clive Andrews Mat Hobbis
3. Adoption Cycles Functionality & Performance Time
4. Adoption Cycles Functionality & Performance Time
5. Considerations in Selecting Hardware vs. Software
6. Hardware: Extreme Performance
7. Hardware: Predictable Behavior
8. Hardware: Massive Scale
9. Hardware: 24x7x365 Reliability
10. Hardware: Cost
11. Software: Versatility
12. Hardware Products are Widely Available and Simple to Deploy
13. Software on Servers Enterprise Appliances o Easier Operation o Lower TCO o Higher Performance Database Storage Web Infrastructure Messaging Middleware IP Routing Applying the Appliance Advantage to Middleware
14. Don’t Make Headlines “India stock exchange flash crash erases US$58 Billion” October 2012 “IT leaders face pay cut after TSE outage” August 2012 “Facebook crashes the Nasdaq” December 2012
15. Application Evolution 15 CONFIDENTIAL
16. 16 Event Driven Architecture (i) • Need to be Agile. • Increased Regulation . Audit, “Real-Time” Global Risk and P&L • Drive EDA for scale and resilience – drives message bus requirements • Bus Latencies and Throughput important Trade Bus Trade Bus Monitor /Staging Bus Monitor / Staging Bus OMS Sub Post Trade Dist Svc DB Persist OMS Sub Post Trade Dist Svc DB Persist Risk P&L
17. 17 CONFIDENIAL Event Driven Architecture (ii) • Co-Locate Processes where Latency is key • Shared Memory IPC within host (Same API) • Non “on host” components also need Low Latency Connections. • Lower Latency requirements of Staging area allow message batching – Turn Message Rate problem into a Bandwidth Problem • Need High Availability and recovery options • Need Disaster Recovery options Trade Bus OMS Sub Post Trade Dist Svc DB Persist Crossing Engine SOR Physical Host Shm Q
18. Product Architecture 18 CONFIDENTIAL
19. 19 CONFIDENTIAL Networked Architecture • Hardware-based middleware overlay for IP networks • All Message QoS in one Appliance – Reliable/Persistent/Web Streaming • WAN Optimisation and Compression • Comprehensive Statistics and Monitoring
20. 20 CONFIDENTIAL Modular Addition of Functionality Data Plane Capabilities embedded in FPGAs and network processors, added via modular architecture - Build to suit - Scale within footprint - Easy upgrades Control Plane Administration, subscriptions and stats collection never impact performance High-Speed Interconnect (10 blades in 3260, 5 blades in 3230) Solace Blades (PCIe Cards)
21. Reliable Messaging • Pure hardware solution – No operating system – No context switching – No interrupts – No data copies • 10 million messages/second – Can be any combination, e.g. 5M in & 5M out, 2M in & 8M out 21 CONFIDENTIAL 22 23 24 26 29 35 25 26 30 32 39 54 0 10 20 30 40 50 60 500K/500K 1M/1M 2M/2M 3M/3M 4M/4M 5M/5M Avg 99.9th Messages per Second Micro- seconds of Latency Bulk Message Rate Message Size (bytes) Message Rate (msgs/sec) User Payload Bandwidth (Mbps) 100 5,930,000 4,744 500 2,080,000 8,320 10GigE Line Rate is the Limit 1,000 1,080,000 8,640 12,000 92,000 8,832 30,000 34,000 8,160
22. CONFIDENTIAL Guaranteed Messaging; Store & Forward Performance Failsafe w/o overhead of persisting every message to disk 200K msgs/sec ingress and 200K msgs/sec egress Latency steady even while recovering when disconnected subscribers reconnect 22 69 69 69 73 79 84 98 90 88 91 99 114 123 154 0 20 40 60 80 100 120 140 160 180 2,000 10,000 25,000 50,000 100,000 125,000 150,000 Avg 99.9th Messages per Second Micro- seconds of Latency 206,400 202,000 157,500 124,400 0 50,000 100,000 150,000 200,000 500 1,000 2,000 4,000 ADB-3 Software Broker Msg Rate (Msg/sce) ADB Message Rates
23. Guaranteed Messaging; Cut Through Persistence Latency 23 CONFIDENTIAL Low, consistent latency for low latency trading applications Can also have store & forward clients for same published message Queues can have low and high priority limits set. During congestion : Reject new orders Process changes to existing orders
24. Steady in Face of Slow Consumers o Latency stays consistent even through disconnection and re- connection of clients o Re-connected subscribers “catch up” without impacting other clients 74 75 89 74 103 113 170 103 0 20 40 60 80 100 120 140 160 180 Pre-Failure Spooling Catchup/Recovery Post-Recovery Avg 99.9th Period of Test Micro- seconds of Latency 24 CONFIDENTIAL
25. IPC Shared Memory Messaging • Single API session for: – Communications between processes on one OS instance – Topic-based pub/sub and request/reply – Any-to-any messaging – Reliable delivery • Applications can block or busy-wait • C API for Linux, Solaris and Windows • Move apps to IPC with no application changes 25 CONFIDENTIAL Core 1 Core 3 Core 2 Core 4 Shared Memory 1 publisher -> 1 subscriber • 2.91 million msgs/sec; 128 byte messages • Average latency 431 nanoseconds 99th percentile 480 nanoseconds 6x6 mesh simulation of fanout/fanin • 46.8 million messages per second • 154.5 gigabits per second
26. Thanks……