BiB 061: Understanding DriveScale Composer Architecture

The following text is a transcript of the audio you can listen to in the player above.

Welcome to Briefings In Brief, an audio digest of IT news and information from the Packet Pushers, including vendor briefings, industry research, and commentary. I’m Ethan Banks, it’s November 28, 2018, and here’s what’s happening. I had a briefing with DriveScale last month.

DriveScale is a composable infrastructure company, with a primary use case of distributed storage. That is, DriveScale software de-couples servers and storage, making it possible to compose infrastructure flexibly from pools of available resources. If that sounds more or less like what you get with public cloud, DriveScale claims to provide “the agility of the public cloud at a fraction of the cost.” That’s a plausible claim, as we’ve all learned that a company doesn’t move to the public cloud to save money.

I’ve written and spoken about DriveScale before, so search a bit if you’d like more detailed background on the company or check out the links in the episode notes. Now let’s jump into the discussion.

• An Introduction To DriveScale
• DriveScale’s Software Composable Infrastructure For Flash

DriveScale–Not A Hardware Company

In this briefing, DriveScale explained the current state of their evolution. One emphasis was that DriveScale wants to be out of the hardware business entirely. They are a software company that will work with a variety of different hardware. DriveScale will still supply some hardware if really needed to fill in gaps their customers might have, but that’s not what they want to be doing long-term.

Demonstrating how keen they are to be out of the hardware game, DriveScale highlighted some of their new industry alliances. For example, DriveScale is now a Dell EMC Tier 1 Enterprise Infrastructure Global Partner. That means Dell sales reps can sell DriveScale directly and get comped for it. That also means that if Dell EMC is an approved vendor in your shop, purchasing DriveScale is relatively easy for you.

DriveScale also announced support for the Cisco UCS S3260 Disk System, a monstrous series of disk chassis you attach to the network at high speed via dual QSFPs.

You get the idea. The more mainstream hardware DriveScale works with, the less DriveScale needs to supply their own hardware to make their solution saddle up a unicorn and ride around.

DriveScale Composer Architecture

Very well. DriveScale is all about the software. And what is the software? Much of the magic happens in DriveScale Composer. The chief value of DriveScale Composer is to compose any compute to any disk or flash, in a scalable way. DriveScale has successfully tested 3,000 nodes and 100,000 disks. 10,000 nodes and 100,000 disks are expected over time.

The system architecture is, of course, secure multi-tenant across the shared infrastructure, as you’d expect for something that scales so large. If you have a vast pool of resources, it’s safe to assume that a variety of discrete user groups would be consuming them. So, multi-tenancy is a necessary box DriveScale is checking here.

If you put multi-tenancy, compute, storage, and composability together, you might be wondering if DriveScale is a full-blown orchestrator, like Kubernetes. They are not. DriveScale Composer is RESTful API driven, and can be integrated with orchestrators. But DriveScale made it clear that they are not trying to orchestrate anyone’s applications themselves. Compose the infrastructure an application will use? Yes. Integrate that composable infrastructure with an orchestrator? Yes. Take on full orchestration duties themselves? No.

In addition to APIs, the DriveScale system leverages agents. These are user level agents that manage the resource pools. Server Agents communicate back to Composer about compute while Adapter Agents communicate back to the Composer about storage. Agent data is paired with LLDP information. This gives Composer a decent sense of the network topology along with server and storage node placement.

With that knowledge, Composer will connect resources assuming that an efficient data path and high performance is mandatory. However, Composer will place data replicas in different zones in the data center to insure data integrity during a hardware failure, along with not sharing resources within a pod where a shared fault domain would result.

In other words, DriveScale Composer’s default resource allocation will make sure you don’t shoot yourself in a foot. That said, you CAN take the safeties off and shoot yourself in the foot if you really, really want to. But left to itself, Composer does it’s best to make sure your application doesn’t fail unexpectedly.

In Conclusion

If a public cloud level of infrastructure flexibility for your on-premises compute and storage is interesting to you, search for drivescale and tech field day on YouTube or visit drivescale.com.

That was Briefings in Brief from the Packet Pushers. For more IT podcasts, blogs and news created for engineers, visit packetpushers.net where you can subscribe for free. And for even more great information, become a member at ignition.packetpushers.net.