Белая книга для Cisco Cisco Nexus 5010 Switch

White Paper

Page 1 of 9

Cisco Nexus 5020 Switch Performance in Market-Data and Back-
Office Data Delivery Environments

Overview

In the capital markets industry today, significant growth in the levels of data traffic for pre- and post-trade information

processing has become the norm. Addressing this continued and accelerating growth is a major challenge and

concern for all involved. Those involved are looking at many technologies to stay on top of this accelerating growth:

higher-speed processors, more memory, larger-scale environments, and more recently, high-bandwidth and low-

latency 10 Gigabit Ethernet switching technologies.

The move to these higher-bandwidth technologies is not without its own challenges. In-host protocol processing

continues to be a significant challenge and an impediment to increased performance as it accounts for anywhere

between 70 and 90 percent of the end-to-end application latency. Fortunately this problem is being addressed

through the offerings of offload technologies such as TCP/IP offload engines, user-space communications libraries,

and Remote Direct Memory Access (RDMA) capabilities in a number of 10 Gigabit Ethernet host adapters.

As the host-side performance bottlenecks are addressed, the capabilities and effects of the underlying network fabric

become far more evident. Forwarding methodologies, buffering architectures, ingress and egress queuing

processes, transceiver-to-transceiver throughput and latencies levels, and jitter become significantly more apparent.

With the recent announcement of the Cisco Nexus™ 5000 Series 10 Gigabit Ethernet switches, a new generation of

reliable, lossless, and deterministic performance capabilities can now provide the next level of performance in highly

demanding market-data environments.

This document describes the test results from a proof-of-concept evaluation of the Cisco

Nexus 5020 Switch at a

world-leading capital market company. The goals of the testing were to detect:

●

Increases in throughput and reductions in data load and unload times or data latency

●

Increases in the rate of update messages per second in a multicast market-data environment

●

Throughput, latency, and loss levels across the fabric

●

Reductions in data load and unload times

The findings were as follows:

●

Data delivery time for single-threaded data transfers of a 5-GB data set between a single host and a NetApp

OnTap GX system decreased by 88 percent. Times for multithreaded and multipathed data transfers

increased by a factor of 10.

●

Wombat Latency Busters Messaging (LBM) multicast-based market data reached 1 million update messages

per second at the subscriber for a two-publisher-to-one-subscriber test scenario in comparison to fewer than

200,000 messages per second over Gigabit Ethernet.

In-host microbenchmarks for 10 Gigabit Ethernet were as follows:

●

Line rate was achieved at all packet sizes for both unidirectional and bidirectional traffic flows.

●

Transceiver-to-transceiver latencies maintained a constant rate of 3.2 microseconds for all packet sizes

tested.