Professional Documents
Culture Documents
Data Fabric
Data Fabric
Data Fabric™
www.nyx.com
©2010 NYSE Euronext
All Rights Reserved
PRODUCT DESCRIPTION
Data Fabric™
A New Approach
NYSE Technologies’ Data Fabric™ enables applications to leverage the CPU’s multiple
cores, without needing complex multi-threading code changes. Featuring both LDMA
Key Features
and RDMA capabilities, it seamlessly scales data intensive applications across multiple • Breakthrough application performance:
CPU-cores, as well as multiple machines. -O
ver 8+ million 200 byte messages
per second, with nanosecond
Shared Memory – No Retransmissions latency when running one publisher
Data Fabric incorporates the revolutionary approach of a publish-and-subscribe, application and five subscriber
topic-based, application messaging platform that leverages Shared Memory Access. applications on a single, 8 core Intel
This unique design eliminates retransmission requests and multicast storms that occur Xeon 5500 server using Data Fabric’s
on IP platforms when there are network issues, or when a subscriber is unable to process Local Direct Memory transport
updates at the same rate as the publisher. By leveraging RDMA hardware acceleration -O
ver 1.6 million 200 byte messages
to handle message delivery, Data Fabric also provides protection for the publisher from per second, with single-digit
misbehaving subscribers. microsecond latency between two
applications on two 8-core Intel Xeon
Flexibility and Choice 5500 servers, using Data Fabric’s
Data Fabric supports four message transportation types: LDMA; RDMA over 10Gigabit Infiniband Remote Direct Memory
Ethernet; RDMA over InfiniBand; and direct TCP sockets. By employing the Data Fabric transport
Daemon, clients can ‘bridge’ all of these transport types and filter traffic flowing between
• Flexible deployment options: LDMA,
them, both locally and over a WAN. This flexibility allows architects and administrators
RDMA 10GigE, RDMA Infiniband, TCP
to choose the ideal interconnection solution without requiring any code changes, and
ensures the best performance possible regardless of the deployment choice. • Simplified application design and
development: One API for all transports
Each transport method addresses a different performance goal:
• LDMA – trading venue co-location and other high latency sensitive applications • Predictable business application
performance: ‘Flat’ latency curve
• RDMA – latency sensitive applications requiring processing across multiple servers
• Greater data center efficiency:
• TCP – efficient enterprise fan-out to servers and desktops where only traditional Reduced CPU utilization
Ethernet is available
TCP
Copyright 2010 NYSE Technologies, Inc. All rights reserved. All other product and company names and marks mentioned in this document are the property of their respective owners and are mentioned for identification purposes only. Measurements
are based upon independent tests; results may vary. NYSE Technologies does not guaranty any savings or level of performance.
TR/L5715/100105/300