Main content

Salesforce 'percs' up Heroku performance in Europe

Phil Wainewright Profile picture for user pwainewright December 3, 2014
Summary:

Salesforce adds a high-performance option to its Heroku web app platform in Europe
Offers more consistent latency across the user base when running large-scale web apps

Adam Gross DF14 face
Adam Gross, Heroku

Heroku, Salesforce's platform-as-a-service play for developers, extends its high-scaling Performance Dynos product to European customers today.

Performance Dynos are a high-end version of Heroku's standard containers for running web apps or processes, with twelve times as much memory and additional compute resources. This allows applications to run significantly faster and with more consistent response times — boosting a measure known to specialists as 'perc 95' (more on that below).

The offering will help customers meet quality of service guarantees and performance standards at scale, VP of product Adam Gross told me in a pre-briefing this week.

Our Performance Dynos product is really about our larger customers meeting the highest demands of scale and performance.

It turns out this isn't just a startup problem and this isn't just a San Francisco problem.

Europe catching up

Heroku has seen 100% growth in Europe during calendar 2014, he told me, with big users including Toyota Motor Europe, UK-based utility Severn Trent Water and Swedish broadcaster TV4. The historic two-to-three year gap in technology adoption between the US and Europe is starting to disappear, he added.

I think our European customers are some of the most sophisticated customers we have. The delta, if it hasn't gone away entirely, it's remarkably close.

You have enterprises that are more focused on their customer experience — and those that are more customer oriented are more technically sophisticated, regardless of where in the world they are.

There are three broad use cases for the Performance Dynos offering, he explained.

  • Ecommerce. "Performance Dynos are a great product for Christmas," said Gross. As well as retailers, other examples in this category include Toyota, which although it doesn't close transactions online, allows customers to configure their choice of car.
  • Customer engagement. This is a broad category covering companies that build applications to augment their traditional services and offerings. An example here is shopping mall operator Westfield, which not only shows sales and promotions but also allows shoppers to pre-order food and cinema and event tickets to enhance their visit.
  • Media. Companies like TV4, and Urban Dictionary in the US, which deliver large volumes of video using the platform.

Long tail latency

In all of these and other use cases, the new offering helps address a concern that operators of these online services know as tail latency, a concept originally identified by engineers at Google.

Heroku performance metrics dashboard
If you imagine a distribution curve of the response times experienced by users of a web application, there will be outliers that have a very different experience from the median. Tail latency is the slowdown experienced by those at the long tail of the distribution curve — perhaps the last five percent or one percent of users.

This matters a lot when you operate at scale, explained Gross.

Averages can be deceiving. If one out of twenty users is having a bad experience, the net perception of your application is going to be pretty low. When you're at scale, five out of a hundred turns out to be a lot of users.

Large scale web application operators therefore look at measures known as perc 95 and perc 99 — the performance experienced at the 95th and 99th percentile extremes. Heroku's metrics dashboard (pictured) includes both the median response time and the perc 95, said Gross.

The extra memory and resources available in the Performance Dynos help reduce tail latency by providing more consistent performance at high volumes of usage. Bringing this option into the European region — Heroku operates on Amazon Web Services infrastructure — will also reduce any latency effect of not having to send traffic across the Atlantic. TV4 had been finding as much as 150ms faster response times running in a European datacenter compared to the US, said Gross, although this would depend on the specific application and is separate from the tail latency effect. Being able to keep data in Europe was another factor for customers, said Gross.

We always knew customers wanted to be in Europe for latency and data residency reasons. What made us want to bring this to Europe was the sophistication and scale of our European customers.

My take

An interesting insight  into a new metric that anyone delivering online services and applications at scale needs to be familiar with.

Meanwhile, it seems consumers in Europe are just as demanding as consumers anywhere else and here's evidence that the lag in technology adoption that used to exist between Europe and the US is therefore shrinking fast.

Disclosure: Salesforce is a diginomica premier partner.

Image credits: Courtesy of Salesforce.

Loading
A grey colored placeholder image