Introducing Improved Performance Dynos

Last year, we launched the original Performance dyno, designed to support the largest apps running at-scale with more consistent service and faster response times. Today, with the goal of continuing to support our fast growing customers with more flexibility to choose the type of dynos best for their applications, we are excited to announce improvements to our performance dyno lineup:

  • Performance-L — an improved and more powerful version of the existing Performance dyno, renamed the Performance-L dyno
  • Performance-M — an entirely new dyno and smaller sibling to the Performance-L dyno

The Performance-L dyno now has 14GB of RAM, 133% more RAM than the previous version, answering a request from many of our largest customers. We’ve also upgraded the underlying virtual compute instance for Performance-L dynos to the latest generation, ensuring better performance for the largest, most demanding high traffic applications. We’ve made both of these improvements without changing the price.

The new Performance-M dyno comes with 2.5GB of RAM — two and a half times more than the top-of-the-line Standard-2X dyno —coupled with more compute resources. Just like the Performance-L dyno, Performance-M dynos execute in LXC containers, each of which fully occupies the underlying virtual compute instance. Each Performance-M dyno is dedicated to a single customer, which means that customer gets the benefit of an entire virtual machine with that dyno. This architecture translates to highly consistent and faster response times, including reduced 99th percentile latencies, for high traffic applications.

By improving and expanding the performance line of dynos, customers now have much more flexibility to choose both the type and number of dynos best for their applications. At the high end, Performance-L dynos are more powerful and performant than ever. The new Performance-M dyno type gives customers a middle path between Standard-2X and Performance-L, and suits intermediate use cases where a smaller number of Performance-M dynos will deliver better results than a larger number of Standard-2X dynos.

The New Performance-M Dynos

Below the surface, a Performance-M dyno runs in an LXC container that fully occupies an entire current generation virtual compute instance, with reduced dyno boot times and latency. The result is that apps running on these instances will have more CPU time spent executing the app’s processes, and less time on system processes. Overall, these upgrades allow more customer applications to achieve the consistent and predictable performance that customers have long enjoyed from our performance dyno type.

Performance-M dynos are available immediately, and applications can be migrated in seconds. In both the US and EU regions, these dynos can be used for $250 per month, prorated to the second, and provisioned through the Heroku Dashboard or Toolbelt.

$ heroku dyno:type web=performance-m

An Upgrade to the Performance Dyno

The Performance dynos that our customers have relied upon for the past year have been upgraded and renamed to Performance-L. The available memory for each Performance-L dyno has been increased from 6 GB to 14 GB and the underlying hardware has also been upgraded to also include SSDs. These dynos will continue to cost $500 per dyno per month, prorated to the second.

The full lineup of Professional dynos are:

dyno type memory compute price
standard-1x 512MB 1-4X $25/mo
standard-2x 1GB 2-8X $50/mo
performance-m 2.5GB 12X $250/mo
performance-l 14GB 50X $500/mo

During tests, our customers have seen, on average, a 20% to 30% improvement in response time after moving from the previous generation performance dyno to the new Performance-L dyno. All applications currently running on performance dynos have been automatically upgraded to the new offering. New customers interested in using these dynos for their applications can get started today.

$ heroku dyno:type web=performance-l


We strongly believe that our customers will benefit from the improved performance offered by Performance-M and Performance-L dynos. We welcome your feedback — let us know what you think.

Browse the blog archives, subscribe to the full-text feed, or visit the engineering blog.