26 By Wally MacDermid, vice president of strategic alliances, Scality “Object storage performance is becoming increasingly crucial as unstructured data stores offer a new way to address the processing requirements of modern applications like AI and analytics. However, throughput figures received in real-world scenarios may not always match up to the claims from your storage vendor.” The above excerpt is from Object Storage Performance – Your Mileage May Vary, a wonderfully insightful blog written by Chris Evans in 2021. I recently revisited it because there are various “world’s fastest object storage” claims being made by vendors — notably, in the open-source space — and, frankly, I’m confused. Why? Because Scality keeps outperforming these vendors in real-world proof of concepts. I asked some Scality engineers to help me understand the difference between the marketing claims made by other vendors and what we are seeing in the real world. Their answer was quite enlightening: “Not all object vendors are actually providing durable storage — they just provide an S3 API layer.” Wait, I thought, does that mean that an “object storage” vendor can only be providing “object”? As it turns out, yes. And that’s what reminded me of Chris’ blog. Using cars as a lens to look at storage performance testing Chris makes an apt analogy between storage performance testing and how consumers evaluate and think about cars, specifically in terms of kilometers/miles per gallon. Does anyone really believe the “58 mpg city and 53 mpg highway” sticker that the dealership slaps on a car’s window? I think most people understand that advertised fuel efficiency is primarily based on theoretical ideal conditions and that, in actuality, driving at different speeds on different road types will deliver different efficiency numbers. In his blog, Chris outlines six factors that influence the difference between lab performance and real-world observations. They’re important for enterprises to keep in mind when considering which object storage solution to deploy in their data centers. To extend the car analogy to the marketing hype of “world’s fastest object storage” claims, imagine a car with only an engine — no doors, seats, safety belts, or airbags. Surely this car would be lighter and faster than a “complete” car that had all the safety features consumers actually want, wouldn’t it? Well, it turns out that some object storage is really a lot more “object” than it is “storage” — and it’s more like an empty car frame with an engine dumped into it. How safe is this car? Unlike other object storage vendors, Scality offers performance without sacrificing safety Safety — availability and durability — for object storage systems is delivered through a series of technologies such as erasure coding, data replication, failure detection and automated self-healing. Scality’s object storage solutions offer all of these capabilities (and more) to deliver 14 nines of data durability and guaranteed availability — which is much more than even Amazon Web Services can deliver with its S3 storage service. On top of the storage services Scality provides, our products offer protocol layers or “connectors.” Scality supports the Amazon S3 API, NFS, SMB, and other protocols. Cutting through the hype: Take “the world’s fastest” claims with a grain of salt The “world’s fastest object storage” claims you see from other vendors aren’t based on tests of a fully enterprise-ready object storage solution (i.e., a car with all its safety features) — they’ve tested the shell of the car. Scality, on the other hand, offers fast performance with all of the required data protection capabilities. And because petabyte-scale storage systems always have a failure somewhere (e.g., a failed disk), Scality allows our customers to maintain performance while also running mission-critical applications with guaranteed availability. Imagine being able to drive a super fast car with the most advanced safety features — all while changing the oil and tires without ever stopping. Now that’s worth shouting about.
