Vertica in Eon Mode Depot – Where Did You Go?

Posted October 3, 2019 by Joy King, VP Product & GTM Strategy, Vertica

Bullet train rendering with Vertica on side, and zooming effect with "Depot" sign behind

A surprising and exciting advantage with Vertica in Eon Mode for Pure Storage FlashBlade on premise

Separating compute from storage is such a hot topic these days. “Instance types” and “S3 Object Storage” are core to the language of the public clouds (unless you’re Azure, then substitute S3 with Blob but leave the Object Storage part). This next-generation architecture enables so many operational and economic advantages – workload isolation, elastic scaling, hibernate and revive, and so much more. All of this is built on a very critical premise – the network that connects the compute nodes with the storage location is fast and reliable and consistent. And that’s where black and white becomes gray.

The Growing Need for a Separation of Compute Storage Architecture
In contrast, Vertica in Enterprise Mode is built to run on servers with tightly coupled storage. That means that the nodes in any Vertica cluster must have enough storage collectively to store all the data in the Vertica database. If our customers need more data, they need more servers, regardless of whether they need the compute power that comes with those servers. Servers with embedded hard drives means all the data is snuggled up close to the CPUs and the concept of “network latency” isn’t part of the discussion. This is the architecture that has delivered the performance at scale to our customers for more than a decade. So how do we get from this architecture to the separation trend above?

Introducing Vertica in Eon Mode on AWS
In May 2018, Vertica announced Eon Mode, a new option for deploying Vertica on Amazon Web Services (AWS). Sure, AWS provides servers with tightly coupled storage, and many of our Vertica customers leverage Vertica in Enterprise Mode on AWS. But we saw the need to ensure that the flexibility and operational efficiencies that come with the separation of compute from storage was available to our Vertica customers as well, so we created Eon Mode.

Addressing the Limitations of AWS S3 Object Store with the Vertica Depot
Vertica in Eon Mode doesn’t expect servers with embedded hard drives. Instead, it assumes that compute nodes will be connected via network links to the communal storage location. We know that AWS and other cloud providers make commitments on network latency (or lack thereof) but we also knew that we had to protect our customers, especially because we support the world’s most data-driven (and demanding!) customers who can’t compromise on speed. The AWS S3 object store just wasn’t fast enough. That’s when the Depot was born.
The Depot is a function within Vertica in Eon Mode that protects our customers from the risk of variable network speeds, storage response times, and data loading performance. The Depot is often referred to as a form of “cache,” but let’s be clear – the Depot is data storage that is tightly coupled within the compute node. It’s not nearly the same size (well, not usually) as traditional servers but it’s still an embedded hard drive. Vertica in Eon Mode uses the Depot to store “hot data” because this eliminates any network latency that could impact query performance. Vertica will automatically query data stored in communal storage that is not resident in the Depot, but there is an expectation of slower response time. Understand that the Depot maintains an ephemeral copy (lasting for a very short time, in other words); the durable copy is always in S3. The Depot was, until recently, set as “On” for all Vertica in Eon Mode deployments. But then came our support for the Pure Storage FlashBlade.

Bringing Cloud Architecture Down to Earth – Vertica in Eon Mode for Pure Storage
Two weeks ago at the Pure Accelerate Conference in Austin, Texas, I had an epiphany, thanks to AT&T. As everyone knows (if you don’t, don’t tell me – my blood pressure can’t handle it!), we announced the first and the only analytics database platform that separates compute from storage for on-premise data centers – Vertica in Eon Mode for Pure Storage FlashBlade. I was very excited to proudly proclaim that Vertica was bringing cloud architecture down to earth. But I assumed that Vertica in Eon Mode on AWS was the same as Vertica in Eon Mode for Pure Storage FlashBlade. Well, I’m very excited to tell you why I was wrong!

The Benefits of Turning off the Vertica Depot in On-Premises Environments
In an on-premise data center, with Pure Storage FlashBlade, the need for Depot protection is dramatically reduced, and in some cases, eliminated. AT&T shared with an audience at Pure Accelerate that it has turned off the Depot. Why? Variable network latency is no longer a risk. Slow object storage response time is no longer a risk. Data loading network performance is no longer a risk. And, compute nodes with smaller embedded disks are less expensive. So now, we have all the benefits of cloud architecture separating compute and storage without the need for a Depot to ensure the highest performance levels.
Vertica in Eon Mode with Depot On and off
But, as Always, the Choice is Yours for Even Greater Performance
Let’s be clear. Some amount of local disk is required for temp tables and other functions. And a Depot still improves performance marginally. So completely eliminating local disk is not part of this conversation. But, the ability to turn off Depot for an on-premise deployment with Pure Storage FlashBlade means one less administrative step in sizing, managing, and maintaining a Vertica in Eon Mode database with all the flexibility, scalability, and agility that you get from separating compute and storage. And did I mention that those servers are cheaper, too? Did I mention that the performance reliability is still better because queries that overflow to the Depot in AWS can be impacted but without the need for Depot on premise with Pure, and that there is no overflow?
Don’t underestimate the power of “the first, the only analytics database that separates compute from storage for on-premise data centers.” And remember that we always provide our customers with the greatest range of choice to manage their analytical workloads – and never lock you in.

Related posts:
How to Access Flexible, Cloud-Optimized Analytics on AWS
Vertica’s Architectural Direction – Separation of compute and storage
Announcing Vertica in Eon Mode for Amazon Linux 2, now available in AWS Marketplace
Exploring Vertica in Eon Mode for Dynamic Workloads
Introducing an Industry First – Vertica in Eon Mode for Pure Storage