HPE Ships Container Platform with Kubernetes and MapR File System on Board
BlueData had been developing its own container management system since 2012, when two former VMware engineers, Tom Phelan and Kumar Sreekanti founded the Santa Clara firm. By combining a KVM hypervisor, an Open Stack cloud management system, and a Red Hat Linux OS, BlueData’s EPIC platform was able to dramatically simplify the deployment of Hadoop and Spark clusters.
“You may recall that…we had our own container orchestrator,” Phelan told Datanami last week. “We wrote one because Kubernetes did not yet exist. So we watched Docker Swarm container orchestrator come and go. We watched Mesos come and go. But we saw that the Kubernetes container orchestra had become the de facto standard API for the industry, and so as part of the acquisition of BlueData into HPE….we spent the last year enhancing the HPE Container Platform to support the Kubernetes container orchestrator.”
Now that it’s GA, the HPE Container Platform becomes the de facto standard cloud-native container runtime for Hewlett Packard Enterprise. What’s unique about HPE Container Platform is that it was designed to run containers directly on bare metal, thereby bypassing the overhead that’s associated with running containers inside of virtual machines (although it can also run containers in VMs, if the customer wants).
To address the “noisy neighbor” challenge that VMs help solve, HPE has tapped into the Linux cgroup scheduling shares to develop a quality of service component to limit the ability of other containers from consuming more than their share of CPU and memory. It’s also restricting the ability of applications running in the containers to use root access, which addresses the security aspect of why some organizations run containers applications in VMs.
HPE Container Platform currently is certified only for Docker containers, but support for other Open Compute Initiative (OCI)-compliant containers should be available soon. By the end of the year, it will also be available on HPE’s Green Lake private cloud offering, at which point it will be sold on a OpEx basis. Currently, it’s being sold on a CapEx basis.
In addition to adding Kubernetes and changing the name from EPIC to HPE Container Platform, the new platform also includes the MapR File system. That further extends the capability of the platform to handle more workloads than the big data and ML workloads that has been BlueData’s forte, according to Phelan, who is the chief architect at BlueData and a fellow in HPE’s Big Data and Storage organization.,
“Once we had the opportunity to support a real file system like MapR–which has support for multiple file system protocols, not just HDFS but also POSIX NFS and object store–it provided real value to integrate that into the platform so that our non-big data applications, which more typically use the POSIX file system protocol, can access the data from that file system,” Phelan said.
Adding MapR to the mix provides a venue for running a variety of different applications at scale in a Kubernetes-managed container system, says Jason Schroedl, vice president of marketing for BlueData.
“We have flexibility for customers, whatever compute service they want to run, whether it’s Hadoop, Spark, or Kafka,” Schroedl said. “That continues to be our got to market and sweet spot. What’s new is we’re expanding the scope of the applications and use case and workload that can run on platform, above and beyond AI, machine learning, and data analytics extending across database and a wide range of other data-intensive apps, like IoT and edge analytics, as well as other workloads that today can be containerized and run on Kubernetes, like CI/CD pipelines, DevOps use case, and broader application modernization use cases.”
Hadoop continues to be a big focus for BlueData, and the HPE Container Platform ships with certified copies of Hadoop distributions from Cloudera, the sole remaining pure-play Hadoop distributor. HPE does not consider itself to be a Hadoop distributor. But it is committed to aspects of the MapR roadmap, especially as it pertains to the separation of compute and storage and the ability for multiple applications to access the same pool of data, Schroedl said.
“There is very tight convergence and alignment with the MapR roadmap and BlueData roadmap,” he said. “We’re working very closely with all of them in our roadmap going forward for MapR as part of HPE and this new HPE Container Platform, leveraging technology from BlueData and MapR.”
HPE continues to work with the MapR customers to determine what they want. The company has renewed maintenance agreements with the majority of MapR customers, Schroedl said. It also issued an interim release of the MapR platform, version 6.3, late last year. But aside from the inclusion of the MapR File System into the new HPE Container Platform, there do not appear to be any solid plans regarding the future of the MapR platform.
“In terms of the future development and maintenance of the MapR platform itself, it’s still too early,” Phelan said. “We’re still working with the customer base to see what components of that we’ll bring forward and perhaps which components we won’t.”
If there’s one thing that everybody can agree on, it’s that Kubernetes is the future for managing containers. MapR was ahead of the curve relative to other Hadoop distributors when it came to Kubernetes. The company had shipped a Kubernetes volume driver in 2018, giving its customers the capability read and write data to and from ephemeral MapR applications running in Kubernetes-managed containers–on-prem, in the cloud, or both–to a persisted data storage layer that survives the coming and going of those containerized apps.
“I think the direction we’re going with containers and Kubnernetes is very much in line with a lot of what we’ve heard from the MapR customer base,” Schroedl said.
The HPE Container Platform supports the three most recent releases of Kubernetes, including 1.17, 1.16, and 1.15. Any application that supports those releases will be able to run on the platform, whether it’s running on-prem on bare metal or in the cloud.
This article originally appeared at sister publication Datanami.