Akamai Rolls Out New “Fast Purge” Solution; Questions Remain About Speed and Scale

For well over a year, customers of Akamai have been complaining about Akamai’s cache invalidation times, which has impacted manifest caching, real-time news feeds, and any time sensitive content. Historically, Akamai purges used to take at least 15 minutes and sometimes, in a couple of really terrible cases, I’ve heard of hours. Competitors like Fastly have been quick to jump on Akamai’s purging limitations and have been winning deals in the market based on Fastly’s ability to purge content within hundreds of milliseconds.

It seems that some CDNs caching strategies are old school and based around the idea that you have little to no real-time control of your caches. You set a TTL and if you need to invalidate, it can’t be mission critical, you just have to wait minutes or hours. So you set your caching strategy by identifying what you can afford to behave like this and then build against that, so your home page, news feeds, api’s, dynamic elements, HLS manifests etc. can’t be cached. Customers tell me that with Fastly, they have turned the caching strategy on its head. They cache everything (except truly uncacheable content like PII or specific to a single user), and then invalidate it as needed. It’s the difference between 90% cache hit rate and 99.999%. New York Times is a classic example of an Akamai customer, which serves all their HTML from origin and only caches images on Akamai because they don’t have this capability.

Akamai has been aware of the major limitations of their platform when it comes to purging content (see their blog post from January) and has been building out a new system, which allows purging as low as a few seconds. That is a dramatic improvement, but content owners have been asking me how widespread Akamai’s new system is, if it is available on their entire platform, or if there’s any rate limiting. Some Akamai customers tell me they are still in the 15 minute purge range. By comparison, when they compare Akamai to Fastly, their entire platform supports instant purge, they don’t rate limit and you can purge the entire cache if you want and it’s all API driven. Fastly customers tell me they have 150ms or less purging capabilities.

So with all these questions out in the market, I had a chance to speak to Akamai about how they are addressing their purging issue and got details from the company on their new platform, made available this week, which looks to address some of their customer’s purging complaints. Akamai’s new “Fast Purge” solution enables Akamai customers to invalidate or delete their content from all of Akamai’s Edge servers via API and UI in “approximately 5 seconds”. With Akamai’s Fast Purge API, the company said their customers “can automate their publishing flow to maximize performance and offload without compromising on freshness.” With this “Hold Til Told” methodology, Akamai customers can now cache semi-dynamic content with long TTLs, and refresh it near instantly as soon as it changes.

Akamai says the Fast Purge UI will complete the roll out process this week to all customers, and is already available to 85% of them. Fast Purge API has been adopted by almost 100 Akamai customers so far and they said it supports a “virtually unlimited throughput of over 100x that of our legacy purge APIs per customer.” Its early adopters include major retailers caching their entire product catalog and major media companies caching news stories and live API feeds for day-to-day operations. In Q1 2017, Akamai says Fast Purge will support purge-by-cpcode and purge-by-content-tag. With Fast Purge by content tag, customers will be able to apply tags to content, and then with one purge-by-tag request, refresh all content containing that specific tag. For example, eCommerce customers will be able to tag search result pages with the SKUs on them, and then when an SKU is out of stock, with one request remove all pages referencing it.

It’s good to see Akamai finally offering a better purging solution in the market, but only customers will determine if what Akamai now offers will fit the bill or not. The keys to instant purge are speed and reliability at scale. Customers say their experience on the “Hold Til Told” approach suggests that you need to trust that purges will happen and they need to be reliable across the world and at scale. If your site depends on being updated in real-time to ensure you don’t sell something you don’t have or provide outdated information the users need to trust it will work. If purges do not happen reliably, it creates mistrust and damages the entire premise of “hold til told”. So customers of any CDN should test purge times under many different conditions and in various regions on the production network to ensure it actually works as advertised. Even more so for Akamai customers, since we don’t know what scale or reliability their new Fast Purge solution has. While Akamai said they now have a 100x increase on the throughput from the legacy system, the old system was so limited that it’s possible that a 100x increase simply isn’t enough and would not meet the needs of many large customers.

Another unanswered question is what Akamai has done to integrate their underlying purging system into major CMS vendors and platforms, so that you get this feature as part of your basic install of the CMS. Akamai has not traditionally worked with the partner ecosystem well and it will be interesting to see how they plan to be on by default in the key CMS and platforms. Competitive CDNs have historically been developer friendly and have well-documented APIs for integrating with other platforms, and that has traditionally been a challenge for Akamai.

On the speed front, it’s good to see Akamai improving, but many businesses would not function with 5 second purges times. For example, customers that have real-time inventory that cannot be oversold. I see this 5 second limitation and the unknown scale and reliability of the system being a huge challenge for Akamai in a market that is truly milliseconds based. It is great they went from minutes to seconds but the performance game is now measured in milliseconds. Scale, reliability and speed are words everyone uses when it comes to delivering content on the web but for purging of content, customers use real-world methodology to measure the impact it has, positive or negative on their business. Customers are the ultimate judge of any new service of feature in the market and at some point, as more look to adopt Akamai’s Fast Purge solution, we’ll find out if 5 seconds is fast enough or not.

If you are a customer of Akamai or any other CDN, I’d be interested to hear from you in the comments section on how fast you need purging to take place.

  • Jeff Zitomer

    Thank you for spreading the word about Fast Purge! I am the Akamai Product Manager for Fast Purge and various other capabilities. We would like to share some additional perspective based on customer feedback.

    Most importantly, there are no questions around speed and scale. We are very proud of Fast Purge being fast and available across Akamai’s entire massive globally distributed network. With Fast Purge on Akamai, customers no longer need to choose between refreshing their content quickly and scale, reliability, availability, performance, and security. They get it all.

    While we are not at liberty to share specific customer details at this time, approximately 100 customers have already adopted the Fast Purge API. During some of the largest online events globally in recent months, customers have put Fast Purge through the gauntlet and we’re excited to report that it met or exceeded all of their business requirements on speed, scale, and reliability.

    Finally, we would like to encourage all of our customers to adopt the Fast Purge API, so that they too can experience purge in seconds at Akamai scale: https://developer.akamai.com/api/purge/ccu/overview.html

    Jeff Zitomer
    Sr. Product Manager
    Akamai Technologies