People are madly registering for VMworld sessions as the 10 year anniversary of the conference opens in less than 2 weeks. I highly recommend this next session in the Extreme Performance Series not only because latency is the next frontier to be conquered in network virtualization, but because of the speakers. Lenin and Haoqiang are two of my go-to network folks in the performance engineering team when I get hard questions and need to dig deep into the details of our stack.
VSVC5596 – Extreme Performance Series: Network Speed Ahead
Lenin Singaravelu , VMware
Haoqiang Zheng , VMware
Performance demands of extremely latency-sensitive applications have long been thought to be incompatible with virtualization. Such applications as distributed in-memory data management, stock trading, and streaming media demand extremely low latency or very low jitter, typically of the order of few tens of microseconds. While virtualization brings the benefits of simplifying IT management and saving costs, the benefits comes with an inherent overhead of abstracting physical hardware and resources and sharing them. These two key features (and sources of benefits) of virtualization, abstraction of physical hardware and sharing of resources are responsible for increased processing time and its variability. vSphere ensures that this overhead induced by virtualization is minimized so that it is not noticeable for a wide range of applications including most business critical applications such as database systems, Web applications, and messaging systems. Extremely latency-sensitive applications, however, may be hurt by the overhead due to its strict latency requirement. In order to support VMs with such strict latency requirements, vSphere has a feature called latency-sensitivity. With this feature enabled, a VM can achieve near-physical performance in latency by removing the major sources of overhead in two ways: it allows VMs 1) to exclusively own physical resources to eliminate contention and 2) to bypass virtualization layers. In this session, we present details of the latency-sensitivity feature and discuss its benefits and limitations.
The draft of this session I’ve seen is packed with experiments, data and some cool new offerings in the vSphere platform. If you’ve ever wanted a deep dive into the network stack and are concerned about network latency or jitter, this session will deep dive into what VMware is doing to address this and just how successful you can be.