Intel's technology for Layer 4 Server Load Balancing

Author: Miguel Lopez

Co-Author: Babu Peddu

Connectivity within the last ten years has increased in ways that were perhaps unfathomable in the early days of the internet. We now are living in an age where streaming, webcasting, and videocalls are expected and relied upon for education, business and entertainment. Consumers increasingly want services that allow them to transfer more data than ever before.  In 2022, world internet traffic is estimated to be approximately 4.8 zettabytes.² By 2026, internet traffic will triple to 15 zettabytes.³ Data center operators will need to add more servers to their infrastructure to serve this growing demand. But this creates a challenge: how can load balancing be managed to address this increased traffic in a manner that is cost-efficient, provides a smooth user experience, and brings optimal revenue to data centers?

Intel technology for layer 4 server load balancing

Intel’s technology for layer 4 server load balancing is targeted to run on server switch white box systems to create an acceleration solution. The solution delivers unique capabilities by leveraging open-source standards like the P4 programming language and cloud-native technologies such as the SONiC open-source networking operating system which provides extensibility to do load balancing. The solution also utilizes high-performance application software components like open-source streaming database Redis and can perform load balancing for IPv4 and IPv6. The solution’s programmability and the separation of control and data planes provides flexibility for customizations, field updates, and upgrades.

Normally, behind a public datacenter IP address, multiple hosts can serve a client request. These requests are distributed evenly across available resources through load balancing. This is typically handled by the server load balancer that chooses a backend server and stores relevant information in memory. All subsequent traffic is directed by the server load balancer to the same backend server. The acceleration solution sits in front of the server load balancer, acting as a cache and providing a fast path for established connections, which avoids going through the server load balancer altogether. The result is a session aware solution that creates faster delivery of the content to the client and improved user experience.  The server switch increases load balancing capacity per Watt as well as by unit of real estate. The saved power budget and real estate is then available to servers generating more revenue. This helps ensure application availability and elastic scale-out of server resources, that leads to overall improvement of data center efficiency.

Acceleration Solution Key Benefits

Most of today’s common solutions to load balancing are either appliance-based or software-based.  The appliance-based solutions are usually inflexible proprietary solutions that come with many features that may not be needed by the customer. This creates limited vendor choice and can constrain the ability to quickly modify load balancing to meet changing customer demands. On the other hand, software-based solutions have greater flexibility and can initially be deployed more easily and affordably. However, these solutions can lead to higher TCO as more long-term investment is required to scale them to Tbps load balancing speeds.

Intel’s acceleration solution provides accelerated load balancing while also overcoming the limitations faced by appliance-based and software-based solutions. The acceleration solution has shown promising potential by providing up to 10x TCO reduction over existing solutions through its capability of handling multi-terabit speeds and 100s of millions of concurrent sessions.⁴ Intel’s acceleration solution operates flexible load balancing that is based on a programmable data plane allowing for support of various load balancing strategies and tunneling or encapsulation options, which in turn offers scalability and flexibility to customers. Additionally, the acceleration solution’s single digit microsecond latency of load balancing operation provides faster access to data, improved performance, and enhances user experience.  Because the acceleration solution is upgradable, it is paving the way for flexibility in vendor choice for data centers.  Lastly, the optionally integrated SYN-flood DDoS attack defense feature provides line-rate performance and near-zero resource consumption.

Intel’s acceleration solution enables your enterprise to serve 100s of millions of concurrent sessions with great response time.  Thanks to the programmable hardware data plane implementation, the solution offers plenty of capacity to accommodate traffic spikes and operate at line rate.  This is an aspect also shared by Alibaba’s Sailfish cloud gateway project which, like Intel’s acceleration solution, was able to manage peak traffic for online shopping festivals in China with multi-Tbps throughput and less latency.⁵

With the world becoming more interconnected and dependent on increasingly data intensive online services and products, data centers are pressed to deliver better results at fast speeds. Data centers will need flexible and cost-efficient solutions to manage load balancing and to be able to meet customer demands. Intel’s technology for layer 4 server load balancing boldly meets this challenge by empowering data centers to fulfill growing customer demand with optimal performance, greater energy efficiency, and reduced costs.

Join us at VMware Explore 2022 held in the US from August 29 through September 1, 2022, at the Moscone Center in San Francisco, CA. We will be demonstrating Intel’s server load balancing technology in The Expo (South Hall – Lower Level), Booth# 901.  We also welcome you to attend our theater presentations by Babu Peddu about the server load balancing technology on Aug. 30 at 3:00 pm, Aug. 31 at 11:00 am, and Sep. 1 at 2:30pm.

Citations

1. https://doi.org/10.1145/3387514.3405885
2. https://wdr2021.worldbank.org/stories/crossing-borders/
3. https://cis.temple.edu/~giorgio/threads_whitepaper.pdfhttps://www.analysysmason.com/contentassets/c85344f3c47b4108ac24e6e919321aa9/analysys_mason_fixed_traffic_forecast_sample_aug2021_rdfi0_rdmb0.pdf
4. https://github.com/iqiyi/dpvs.  TCO estimate based on performance report for DPVS software-based load balancing - - It demonstrates performance of around 2.5M packets per second per core.  Utilizing 40 cores per processor, a dual socket server can thus achieve 200M packets per second. The Intel acceleration solution can do up to 4.8B packets per second, making it possible to achieve the same or better packet forwarding capacity of 20 dual-core servers using a single server switch.
5. https://builders.intel.com/docs/networkbuilders/alibaba-manages-peak-traffic-with-cloud-gateway-1656700556.pdf

Notices and Disclaimers

• Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.  
• Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates.  See backup for configuration details.  No product or component can be absolutely secure. 
• Your costs and results may vary. 
• Intel does not control or audit third-party data.  You should consult other sources to evaluate accuracy.
• Intel technologies may require enabled hardware, software or service activation.
• © Intel Corporation.  Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries.  Other names and brands may be claimed as the property of others.