Virtualization, NFV, NSX-T

DNS is in most cases currently still a 'bare metal' service, often on legacy hardware. New deployments in telecommunications environments are more likely to target a virtualized environment at least, and frequently also NFV, OpenStack or VMWare NSX-T Data Center deployment.

PowerDNS has ample experience in migrating nameservers from bare metal to more flexible environments. Whenever time sensitive services get virtualized there is a legitimate worry if there will be a latency impact. A naively implemented virtual nameserver will indeed deliver a 95th and 99th percentile performance that is worse than a bare metal deployment.

Through experience on a wide variety of platforms we have developed configurations that deliver the best latency on common service provider virtual/VNF platforms.

In addition, virtualization makes it possible to spin up new capacity on demand. This requires strong automation and control plane abilities, which is part of our PowerDNS Platform offering.

Latency

Consumers experience the sum of radio network latency (WiFi, 4G, 5G), last mile & transport to DNS server latency plus finally the latency contribution of the nameserver itself, including that generated by its virtualization platform.

Whenever consumer magazines, consumer organizations or governments measure the performance of an Internet Service Provider, they measure things like page load times, ping times, raw bandwidth speeds but frequently also DNS response times (and error rates).

DNS response times are impacted by many things, including hardware speed, virtualization overhead and cache hit rates. Cache hit rates are meanwhile strongly influenced by server concentration - the more users that use a single server cache, the higher the cache hitrate, the lower the latencies delivered. But this is all only true up to a point - an overloaded server is detrimental to perceived performance.

Key to delivering top latency is therefore concentrating high volumes of traffic on small numbers of CPU cores. And this in turn requires expert configuration of the chosen virtual platform.

Performance

A key challenge is providing guaranteed performance levels, also on virtualized hardware. Providing such guaranteed numbers was not easy on legacy bare metal platforms and has been somewhat of a challenge on virtual platforms.

PowerDNS is pleased to be able to provide QPS and latency numbers for specific hardware situations, enabling customers to provision sufficient amounts of cores to deliver desired performance levels.

Virtualization benefits

Virtualized servers can be more efficient because idle servers (say, for printer or payroll purposes) take up no resources when they are not busy. This enables virtualized platforms to deliver impressive savings.

However, nameservers are never idle and should in fact be as busy as possible. This flies in the face of most virtualization configuration guidance which strongly recommends oversubscribing servers to reap the benefits of virtualization.

Through ample experience migrating 'bare metal' deployments to virtualized equivalents, PowerDNS can be a strong partner in delivering a virtual nameserver platform that delivers real performance.

End-to-end monitoring and reporting

With the growth of traffic levels, congestion can crop up anywhere. Such congestion immediately impacts DNS latencies, something that reflects badly on your operations.

When deployments are virtual, not only can networks be congested, so can virtualization platforms. This exhibits itself as hard to debug 'stolen' time that slows down servers without being visisble from the operating system.

To make sure your DNS platform shines, PowerDNS is pleased to offer end-to-end performance monitoring solutions or services. In short, this delivers high resolution quality assurance from many vantage points. By performing differential measurements, this platform can deliver a verdict on actual perceived network performance as well as help attribute problems to relevant network elements, links or (virtual) servers.

Daily and weekly reporting meanwhile provides insight into average and worst case performance, plus flags time intervals in which configured performance levels were not met.

Control plane

Virtualization makes it possible to spin up new capacity on demand, or spin it down when it is no longer needed. Such capabilities requires a strong control plane to make sure configurations are always deployed/retrieved correctly, and that all active nodes contribute correctly to statistics, streaming telemetry and performance monitoring.

PowerDNS is pleased to deliver a fully automated deployment platform with self-configuring telemetry that can deal with auto-scaling and distributed platforms, while retaining centralised insight.

More information

For more information, please contact us directly so we can discuss how we can virtualize your DNS deployment while retaining peak performance.