As anyone who shares a broadband network with high-capacity users knows, when others are eating away at bandwidth by playing high-definition games, streaming TV shows or live broadcasting over social media, internet performance deteriorates.
The same is true for any shared IT resource. Including Nfield’s survey hosting environment in the Azure cloud. Our acceptable use policy is designed to prevent excessive resource use by some causing performance problems for all. But with the roll-out of our Isolated Interview feature, we aim to completely eliminate the risk of suffering from “noisy neighbors”!
The Noisy Neighbor effect
In a shared hosting environment, an individual survey occupying a disproportionally large amount of processing power will lessen everyone else’s performance. This can happen if a lot of respondents suddenly rush to a survey at the same time, or if a badly scripted questionnaire results in an infinite loop or excessive quota checks.
To prevent this in situations where it can be foreseen, such as the planned launch of a large-scale research project, we ask customers to warn us in advance so we can scale-up capacity as necessary. But errors, such as bad scripts, can overload the system without warning.
Nfield does automatically expand its capability as necessary, but this isn’t instantaneous. In the period before scaling up takes place, online interviewing performance and Nfield manager responsiveness slow down. Respondents can get frustrated and the market research team’s job becomes more difficult, as everyone suffers the consequences of having “noisy neighbors”.
Proactive protection and appropriate scale-up
Thanks to new technology, we have succeeded in developing an architecture whereby each individual survey is run in isolation (survey-level isolation), utilizing its own resources and with the ability to expand these without affecting other surveys.
When an isolated survey reaches its pre-set Nfield threshold, its resources will automatically scale up, without other surveys in the same or different domains being impacted during the moments of heavy loading or scaling. In Nfield design, if a request faces timeout, there is a double fall-back mechanism to keep servicing the request. Initially falling back to the isolated resource, then to the shared resource.
Expanding in this way at survey level is also more economic than doing so at deployment level (as in the noisy neighbor scenario).
By keeping every survey accountable within its own realm, Isolated Interviewing
- eliminates noisy neighbor effect
- scales more economically
A record-breaking case
Isolated Interviewing doesn’t just provide protection from noisy neighbors. It also enables heavy users to scale up faster and cheaper than when in a shared environment. A shining example is a recent project in which 2.36 million requests were handled in a four-day period, peaking at 5,000 requests per minute. This also saw us smash right through the 100,000 successfully completed Nfield interviews in 24 hours barrier for the first time!
Despite this extremely heavy load, processing time remained reasonably fast, with 99% of requests processed within 549ms, and no other customers impacted. In addition to performance speed, the Isolated Interviewing feature provides a more stable, risk-free environment for everyone.
Global roll-out
All domains within our America deployment environment already have Isolated Interviewing enabled. However, Isolated Interviewing only comes into effect in newly-created surveys. Existing surveys remain on shared resources.
Over time, we’ll be carefully rolling Isolated Interviewing out to more domains in other deployment environments. And, of course, we’ll continue to invest in this feature to further increase its capabilities to accommodate our customers’ hunger for high-volume projects.
