I'm Jeff Costa, product manager at Akamai for EdgeKV (and a few other things). As a CDN running hundreds of thousands of servers in 135 countries, we know something about caching, networks, and distributed computing. After we launched EdgeWorkers, our serverless compute product, we quickly realized that where data was located was critical. If you run JavaScript functions at the edge, but have to reach back to a cloud database or a single datacenter, the round-trip latency could negate the performance benefit. EdgeKV was built to solve that problem.
We decided to start with a key-value store layered into the CDN model. The focus was making database reads as fast as cached object reads, while writing data quickly. We built a new cache layer for EdgeKV to ensure that when EdgeWorker code needs data (ex: to make a decision inside a function), it pulls that data from the adjacent data cache in ms. For writes, we landed on an eventual consistency model as it provides reliability via redundancy. When a write is made, we send it to multiple nodes in our persistent backend, then update any copies of it that exist on edge servers - without impacting read performance. This means that when you write a value, it may return stale data for a short period of time when you attempt to read it - but will eventually return the latest value in subsequent reads. The period of time between write and read convergence is called the "inconsistency window," and our goal is complete global data writes within 10 seconds. Ten seconds.
Think about what apps and data can benefit from living closer to your customers. Beta signup form, GitHub, and docs listed below. EKV is beta, so expect a few warts. We're excited to hear your questions and feedback on what we've built, so please reach out or leave a comment below. Thanks for reading!
I'm Jeff Costa, product manager at Akamai for EdgeKV (and a few other things). As a CDN running hundreds of thousands of servers in 135 countries, we know something about caching, networks, and distributed computing. After we launched EdgeWorkers, our serverless compute product, we quickly realized that where data was located was critical. If you run JavaScript functions at the edge, but have to reach back to a cloud database or a single datacenter, the round-trip latency could negate the performance benefit. EdgeKV was built to solve that problem.
We decided to start with a key-value store layered into the CDN model. The focus was making database reads as fast as cached object reads, while writing data quickly. We built a new cache layer for EdgeKV to ensure that when EdgeWorker code needs data (ex: to make a decision inside a function), it pulls that data from the adjacent data cache in ms. For writes, we landed on an eventual consistency model as it provides reliability via redundancy. When a write is made, we send it to multiple nodes in our persistent backend, then update any copies of it that exist on edge servers - without impacting read performance. This means that when you write a value, it may return stale data for a short period of time when you attempt to read it - but will eventually return the latest value in subsequent reads. The period of time between write and read convergence is called the "inconsistency window," and our goal is complete global data writes within 10 seconds. Ten seconds.
Think about what apps and data can benefit from living closer to your customers. Beta signup form, GitHub, and docs listed below. EKV is beta, so expect a few warts. We're excited to hear your questions and feedback on what we've built, so please reach out or leave a comment below. Thanks for reading!
Jeff @cosjef
[0] Beta signup: https://bit.ly/3qBVesV [1] EKV User Guide: https://bit.ly/2OKVLvt [2] GitHub repo: https://bit.ly/3cqblVh