Back To Schedule
Monday, March 22 • 08:40 - 09:10
Handling Elephant Flow on a DPDK-Based Load Balancer

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
With the fast-growing traffic of high definition video, Cloud Gaming, etc., the handling of Heavy Hitters, or Elephant flows, is becoming increasingly important to QoS requirement. Elephant flows is defined as an extremely huge continuous flow over the network link. It has been reported that 1% of flows generates around 90% of total traffic in the data centers. Due to the large amount of traffic of elephant flows, a single CPU core may not be able to handle an elephant flow completely by itself. Thus, to avoid packet drop, an elephant flow needs to be distributed to a few CPU cores for processing. As the ingress gateway of data center, we found it necessary to deploy heavy hitter detection capability in the load balancer to balance traffic processing among multiple cores. In this talk, we introduce our Heavy Hitter handling practice on a DPDK based High-performance L4 Load Balancer - HDSLB. We will demonstrate how we detect the Heavy Hitter and distribute the load to multiple idle worker cores for parallel processing. We prototyped an Intel® Dynamic Load Balancer (DLB) based solution, as well as a rte_ring based pure software solution for traffic distribution and reordering. We compare the performance gains of both solutions.

Presentation Language: Mandarin (Translation services will be provided)
Presentation Slides will be in English

avatar for Hongjun Ni

Hongjun Ni

Technical Leader, Intel
Hongjun Ni has been focusing on Cloud Networking and Network Security. He is FD.io VPP Maintainer, UDPI Project Lead, Sweetcomb Project Lead and NSH_SFC Project Lead. He has fifteen years' rich experience on Cloud Networking, Network Security, SmartNIC and Wireless. He has given 20... Read More →


Yipeng Wang

Software Engineer, Intel
Software Engineer - Intel

Chenmin Sun

Software Engineer, Intel

Monday March 22, 2021 08:40 - 09:10 UTC
Virtual Experience