HPE’s Alletra MP marries storage to cloud and software-defined choices


Replace the ordeal of hardware admin with the insight of the cloud – that’s the benefit promised to customers that deploy HPE’s Alletra MP storage arrays with HPE GreenLake.

Alletra MP, launched in April, is based on a scale-out architecture of Proliant servers chassis and Aruba switches with nodes that can be compute only – i.e. controller – or compute and storage, or even just storage capacity. Array and backup and restore configuration come from a few clicks on a web console that takes care of all settings automatically.

“It’s true that GreenLake was initially a financing offer so rather than buy hardware, the customer could pay for its usage on a monthly basis in the same way they would pay for the cloud,” said Olivier Tant of HPE in France, to LeMagIT.

“Then the offered evolved towards bundles ready to deploy, for SAP, for big data etc,” he said. “The customer told us their needs and sizing requirements, but it was still necessary to deploy systems for administration.

“Today that’s no longer the case. Admins can connect to the Greenlake cloud console and it allows them to supervise all hardware deployments, as well as the cloud services we provide, and to set their consumption thresholds and ceilings. This web console delivers instructions to the hardware and gets metrics back.”

Data Services console abolishes admin chores

“Beyond automation, not having an admin console on-site offers numerous advantages,” said Michel Parent, also of HPE France. “You don’t need to update virtual machines and there’s no more fear that malware will get in and launch an attack.”

“The storage console in cloud console is called data services,” said Tant. “First of all, you’ll find the storage mode to apply to Alletra MP, block or file. For now, an array or a cluster of array nodes can only function in one mode. File mode comes from the installation of Vast Data [an HPE partner] on the control nodes.

“We’re talking about the possible choices for Alletra MP here, but from the console you can deliver all the storage we offer like Scality, Qumulo, Cohesity and others.”

Then there are tools for monitoring performance – including Infosight, which brings predictive monitoring – for updates, for creation of volumes, to set performance profiles, access rights etc.

Behind the one simple click that activates backup, replication to a secondary site and recovery to normal activity are complete solutions that are deployed from a container, such as Zerto version 10, on-the-fly replication software that detects ransomware infections.

There is also GreenLake Backup & Recovery (which is compatible with AWS VMs, SQL databases, and soon Kubernetes clusters). So, backup and replication options can be with any HPE partner or even outsourcing partners that can work via the GreenLake cloud console.  

“When you get your storage array, it queries the Data Services cloud console and downloads everything you’ve defined,” said Tant. “And it will carry out updates as soon as you subscribe to other options.”

Alletra MP: A variable-geometry storage cluster

The design of the new Alletra MP array contrasts with HPE’s preceding 3Par and Primera products. Most obviously, this is that HPE has abandoned use of motherboards using its own ASIC; in their place, each control node uses two classic motherboards, each with an AMD Epyc processor.

“This architecture change is the price to pay for a multi-purpose machine [whence MP],” said Parent. “We have moved to favour the fact of having x86 cores to execute functions that are run in software. The principle is that of software-defined storage.”

In block mode, an Alletra MP array can be deployed in 2U of rackspace with up to 24 TLC NVMe drives, or it is possible to start with eight 4TB SSDs and to add two drives at a time. Beyond 24 drives, one or more nodes are connected to the control node via NVMe/RoCE and 100Gbps Ethernet. HPE uses its Aruba switches to route traffic.

File mode uses Vast Data, not via a container but directly onto the Alletra OS kernel, and is more complex. It needs a minimum of 200TB capacity and requires one storage node with two Aruba switches.

“Alletra MP is an array in which we can put the control nodes and the storage nodes needed for an application,” said Tant. “Some customers want to deploy six control nodes for one storage node while others want one control node for six storage nodes.”

This flexibility is a key point of the new array. The control nodes share access demands from servers and write redundant data on several storage nodes simultaneously.

“The big advantage of this way of working is that we don’t need to use cache,” said Parent. “Not only are the SSDs fast enough, but – above all – there’s no longer the problem of coherence in the case that a controller fails.”



Source link