If your organization is like many others, your data center is facing the growing pains that come with an increasing amount of data from users and applications—and that means your Ethernet networks are under pressure to perform. By identifying and addressing problems with Ethernet latency, you can improve the overall performance of your data center and ensure you’re getting the most from your technology investments.
Why Ethernet Latency Matters
Network latency measures the time it takes for a data packet to travel from one node to another, telling you how quickly your network is able to respond to requests.
In many cases, it’s better to measure latency via packet rate than it is to look at network throughput alone. Ethernet throughput may tell you about your available bandwidth, but it won’t take into account the time it takes for each small piece of data to travel from the transmitting system, across the network, and to the receiving system. These small times can add up when many data packets are being transferred.
Latency affects the interactive use of your network. For example, accessing a website requires multiple requests for text regions, images, menus, and embedded objects. The latency of each request can lead to a noticeable delay.
Interserver communications are also affected by latency. For example, database accesses are often composed of many individual requests that then spawn more targeted requests. These many requests can add up to significant delays.
How to Measure Latency
The easiest way to measure latency is to use the ping command and look at the round trip delay time for a 56-byte packet (an 84-byte packet including IP headers). The time reported by ping is a round-trip time for the minimal packet to get from the sending CPU down its network stack, across the network, up the receiving network stack to the receiving CPU, and then back through the stacks and network to the originating CPU.
Network Latency Solutions
What may be surprising is that problems with latency are often due to software, rather than hardware. Intel Ethernet products are designed with features to help overcome latency issues related to software.
For example, Dynamic Device Personalization (DDP), which is available on select Intel® Ethernet Adapters, improves the predictability of network flows by tagging packets with a specific class. This means the adapter can make decisions about what type of packet is being transmitted and send it to the right place. This leads to improvements in throughput and latency, as well as reduced CPU loads.
Another feature is Application Device Queues (ADQ), which dedicates specific hardware queues and connects optimally to specific threads of execution. The results are improved application predictability, throughput, and latency.
Remote Direct Memory Access (RDMA) is another way to relieve Ethernet overhead for high-speed applications. RDMA enables the network adapter to transfer data directly to and from application memory. This bypasses the host, enabling low-latency, high-throughput communication between applications. RDMA can eliminate parts of the TCP IP stack, which in turn eliminates much of the software latency involved with copying data all the way up the stack.
Intel also provides diagnostic tools to help optimize network configurations, like Intel® Cluster Checker, which can verify the configuration and performance of Linux*-based clusters and suggest remedies. For example, the tool can diagnose high latency associated with Ethernet configuration settings, so that architects can make appropriate adjustments.
Ethernet Component Upgrades
If you’ve already solved for software-related issues and are still experiencing high latency, it may be time to replace outdated network components. Ethernet bottlenecks are a common cause of service delays, especially as processor and storage technologies become more powerful. With the right mix of components, you can get more from your entire data center.
Having network adapters of the right speed can ensure sufficient Ethernet bandwidth to meet your needs. Many data centers are upgrading from Gigabit Ethernet to 10Gb Ethernet. 10GBASE-T offers a simple path to 10Gb Ethernet that’s backward compatible with existing cable. As a market leader in 10GBASE-T, Intel has a number of solutions to support this transition.
Proven Ethernet Solutions
Intel® Ethernet Network Adapters, controllers, and accessories deliver proven performance and cost efficiency. They’ve been tested exhaustively for compatibility, making them a leading choice among customers.
Intel Ethernet products work in tandem with the Intel® Xeon® Scalable platform and the latest Intel® Solid State Drives to maximize the performance of your data center, delivering the best possible service quality to your end users.
Surprisingly, network latency problems are often due to software, rather than hardware. Intel® Ethernet products are designed with features to help overcome these issues.