Cloud ERP provider Plex Systems requires a storage setup that can host hundreds of petabytes, while meeting high thresholds for performance and availability. The software-as-a-service provider is in its final year of a storage transition in which it added NVMe arrays for performance and two additional data centers for high availability.
Plex has been running a cloud for 19 years, since its 2001 inception. It started as a multi-tenant application run through a browser for customers.
“We’ve always been a cloud to manufacturers,” said Todd Weeks, group vice president of cloud operations and chief security officer for Plex. “We’ve been 100% cloud-based to our customers.”
The Plex private cloud runs mainly on Microsoft software, including SQL Server, and Dell EMC storage, including PowerMax all-NVMe arrays.
Scaling out with two new data centers
Weeks said Plex’s capacity from customer data grows from 15% to 25% per year. He said it has more than 200 PB of data for about 700 customers and 2,300 worldwide manufacturing sites, and it processes more than 7 billion transactions a day with 99.998% availability.
“With the growth of our company, we wanted a much better scale-out model, which we have with our two additional data centers,” he said. “Then, we said, ‘Besides just scaling out, is there more we can get out of them around availability, reliability and performance?'”
The company, based in Troy, Mich., has storage spread among data centers in Auburn Hills, Mich.; Grand Rapids, Mich.; Denver; and Dallas. The data centers are split into redundant pairs for failover, with primary storage and backup running at all four.
Weeks said Plex has used a variety of storage arrays, including ones from Dell EMC, Hitachi Vantara and NetApp. Plex is in the final year of a three-year process of migrating all its storage to Dell EMC PowerMax 8000 NVMe arrays and VxBlock converged infrastructure that includes VMAX and XtremIO all-flash arrays.
Two data centers have Dell EMC PowerMax, and the other two use Dell EMC VxBlock storage as mirrored pairs. Backup consists of Dell EMC Avamar software and Data Domain disk appliances.
“If we lose one, we fail over to the other,” Weeks said of the redundant data centers.
The performance advantage
Weeks said switching to the new storage infrastructure provided a “dramatic increase in performance,” both for primary and backup data. Backup restores have gone from hours to less than 30 minutes, and read latency has been at least three times faster, he said. Data reduction has also significantly increased, which is crucial with hundreds of petabytes of data under management.
“The big win we noticed was with PowerMax. We were expecting a 3-to-1 compression advantage from Hitachi storage, and we’ve actually seen a 9-to-1 difference,” he said. “That allows us to scale out more efficiently. We’ve bought ourselves a couple of years of extra growth capacity. We always want to stay ahead of our customers’ needs, and our customers are database-heavy. We’re also making sure we’re a couple of years ahead of where we need to be performance-wise.”
Early all-flash arrays
Plex’s introduction to EMC storage came through XtremIO all-flash arrays. While performance was the main benefit of those early all-flash systems, Weeks said, the XtremIO REST API impressed his team.
“Being able to call into [the API] made it much more configurable,” he said. “Our storage engineers said, ‘This makes my job easier.’ It’s much easier than having to script and do everything yourself. It makes it much easier to implement and deploy.”
Weeks said Plex is reluctant to move data into public clouds because of the fees incurred for data transfers. But it does store machine information gathered from the Plex industrial IoT (IIoT) SaaS product on Microsoft Azure.
“We gather plant floor machine information and tie it into our ERP,” he said. “But we don’t use public clouds for archiving or storage.”
Plex’s IT roadmap includes moving to containerized applications, mainly to support the Plex IIoT service.
“We’re looking now at how we can repackage our application,” he said. “We’re just beginning to go in the direction of microservices and containers.”
Go to Original Article