This was an extraordinary moment for me.
I was standing in front of our demo at the Flash Memory Summit conference in Santa Clara, California, in August 2018, and could not believe how far our Lightbits Labs team had come in such a short time.
Our demo was extraordinary. We were showing the world how Lightbits’ disaggregated NVMe/TCP storage platform can replace direct-attached SSDs in some of the most challenging flash workloads. The demo proved that our platform had average and tail latencies that are as good as or better than directly attached storage latencies. As can be seen in the live demo, the Lightbits disaggregated storage platform achieved read average latency that is 0.97x of the equivalent direct-attached SSD setup, average write latency that is almost 4x better, and tail read and write latencies that are more than 3x better and more than 10x better, respectively.
As I looked around the packed exhibit hall, it struck me how far we had come since Lightbits Labs began in early 2016. At first, our small team worked in the lab. We kept our work very quiet and hush-hush, but we had the drive, excitement, and confidence that we were working on something that would advance the state of the art.
The heat was always on. This is the Israeli way of innovating—fast. Should we do this? Should we do that? How do we build the best hyperscale disaggregated storage platform the world has ever seen?
By August 2017, we were ready for the annual Flash Memory Summit conference in Santa Clara. We worked with Facebook and Intel on our first demo. It was smoking hot and running on the Facebook Open Compute Project (OCP) Lightning platform. At the convention center exhibit hall, we unpacked the equipment and started getting the demo up and running.
At 5:00 am the next day, a hot summer became even hotter, and a small fire broke out at the exhibit hall. The fire thankfully didn’t do much harm, but the automatic sprinklers soaked everything, and the exhibit hall was closed for the entire conference. We had a great demo no one at the summit could see.
The disappointment of 2017 only made the success at this year’s conference all the more sweeter.
In our packed demo space, I witnessed a game changer in the making. Despite going over a standard TCP/IP network, the Lightbits disaggregated NVMe/TCP storage platform can replace direct-attached SSDs while also improving application-level latency. If this sounds like magic, it is, at least in the sense of a sufficiently advanced technology. And soon, very soon, you too will be able to run the Lightbits NVMe/TCP disaggregated storage platform in your cloud data centers.
You will hear a lot more in this blog in the future on the beauty of cloud storage disaggregation and how the NVMe/TCP train, driven by Lightbits, is changing the way cloud data centers operate. Watch this space for updates. And if you’d like to read more about NVMe/TCP in the meantime, see our white paper.
