SparkPost: The Day the DNS Died

Wednesday, March 28, 2018 - 10:55 am11:35 am

Jeremy Blosser, SparkPost

Abstract: 

More than 25% of the world's non-spam email is sent using SparkPost's technology, and our cloud service sends nearly 15 billion messages per month. Running this service in the cloud has provided all the expected benefits of flexibility and scalability, but also unique challenges due to email's inherent nature as a highly stateful, push-oriented service. To support our use case, our service and network utilization models are similarly atypical.

Our DNS needs are particularly extreme. Our infrastructure currently has to support 8,000 DNS queries per second. Two major DNS-related events in early 2017 caused significant delays for our customers and sent us back to the drawing board once again. We recently completed a ground-up DNS tier redesign that includes dedicated VPCs with optimized security groups and ACLs, distribution across tiers and availability zones, resolver tuning and custom configurations, and multiple local caching resolvers per instance.

In this talk, we will discuss our history addressing this challenge, the limitations discovered in our previous approaches, and our current architecture's design and results. Attendees will gain an understanding of what it takes to host a robust DNS service in AWS at a scale beyond what is currently natively supported by AWS' resolver services.

Jeremy Blosser, SparkPost

Jeremy Blosser has worked in systems administration and engineering for over 20 years, and most of that time his focus has been on reliably delivering email and other traffic at scale. He is currently the Principal Operations Engineer at SparkPost, responsible for technical architecture oversight and keeping the cloud service operating and healthy. He lives in Texas with his wife and five kids.

SREcon18 Americas Open Access Videos Sponsored by
Indeed

Open Access Media

USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.

BibTeX
@conference {213060,
author = {Jeremy Blosser},
title = {SparkPost: The Day the {DNS} Died},
year = {2018},
address = {Santa Clara, CA},
publisher = {{USENIX} Association},
}