Forging Industrial Grade, Scale Out Ceph Storage

As tens of millions of Americans who have received their $1,400 Covid relief checks in recent weeks can attest, getting a little financial boost during stressful and difficult times can go a long way in helping to separate themselves from day-to-day survival and turn some of their attention to longer term plans. The same can be said for tech startups with ambitious ideas trying to gain traction in a rapidly evolving IT industry dominated by large, deep-pocketed, established players.

SoftIron launched in 2012 in the software-defined storage SDS field with plans to develop appliances optimized to run the Ceph open-source storage software that simultaneously supports protocols for block, file and object storage. The company in 2017 pulled in $7 million in Series A funding, a decent amount of money that helped it build out its portfolio of HyperDrive appliances powered by Arm-based chips.

The task-optimized appliances, designed and built by SoftIron in Newark, California, are scale-out storage nodes designed to accelerate the performance of Ceph, which in its initial years was aimed at HPC and hyperscale environments but in recent years has begun to find its way into the enterprise datacenters. Enterprises are turning to Ceph to help manage increasingly distributed environments and run advanced workloads like artificial intelligence, analytics and containerized applications, but many don’t have the expertise needed to leverage complex Ceph environments.

SoftIron, which also offers its top-of-rack HyperSwitch switch that runs the SONiC open-source network operating system and is powered by AMD’s Epyc Embedded 300 processors and the Accepherator, which uses an FPGA to accelerate erasure coding for Ceph workloads.

In March 2020, SoftIron announced a $34 million Series B fundraising round, a haul that has allowed it to aggressively grow it workforce and global presence and is helping to accelerate plans as the company looks to add capabilities in the short term and to map out what its long-term future looks like. According to Chief Marketing Officer Andrew Moloney, the number of employees since last year has jumped from about 25 to more than 60 and now has nine offices around the world – three each in the United States (Silicon Valley, San Diego and Dallas), Europe (London, Berlin and Brno, Czech Republic) and Asia/Pacific (Perth, Sydney, and Auckland). Moloney himself joined SoftIron in June 2020.

The company also expanded its offerings, most recently in late March unveiling its HyperSafe enterprise support services for Ceph to make it easier for enterprises to adopt the technology. Ceph has delivered object, block, and file storage in systems that scale into the petabytes and exabyte levels and require high levels of skill to deploy and manage. As enterprises increasingly take on HPC-like workloads, they are turning to flexible technologies like Ceph. SoftIron is using HyperSafe to help with the technology. Other companies like IBM’s Red Hat unit also see opportunities for extending Ceph into enterprise datacenters.

Craig Chadwell, vice president of product at SoftIron, tells The Next Platform that the company saw an opportunity to build and deliver SDS in a way that delivers the capabilities needed for organizations working in environments that span from the datacenter out into the cloud and the edge. The vendor uses standard Ceph and a Linux distribution based on Debian and dubbed SoftIron Linux optimized to support Ceph, then runs it all on hardware designed and manufactured by the vendor and optimized to run Ceph. It’s all wrapped in SoftIron’s integrated management and a support layer.

“You’ve got all of that benefit of a proprietary solution, but it’s still standard Ceph,” Chadwell says. “So if you want to move away from us in the future, you’re not locked in. If we disappear tomorrow, it’s still Ceph, so you can go back to running it on bare metal if that’s what you want to do.”

It’s an alternative to other SDS methods, all of which look for ways to balance cost, performance and complexity, with varying levels of success. As we noted here, another storage startup, Vast Data, is altering its approach, unloading its hardware business and shifting it to partner Avnet and focusing solely on the software that runs in the systems.

SoftIron offers four different HyperDrive systems that are configured based on the need for value, density or performance. The offerings can deliver up to 5 PB in a 42U rack, running at wire speed and consuming at little as 100 watts per node. The company promises to increase density by up to two times and decrease power and cooling costs over other storage, and its HyperDrive storage is available in a pay-as-you-go consumption model. The HyperDrives work with a storage router that work as a services gateway by consolidating virtualization, user shares and other storage technologies into a single platform. This removes the challenge Ceph traditionally presents through its inability to support other file shares and protocols.

HyperSwitch is an appliance that ranges from 1 Gigabit Ethernet to 100GbE and can be used to connect racks and clusters of HyperDrives. Storage Manager is a management tool for HyperDrive appliances, monitoring the health and usage of Ceph storage, manage files shares SMB and NFS, object stores and block devices, like iSCSI and RBD. That combined with the Ceph Button are designed the making managing and troubleshooting Ceph – always a complex task – simpler.

SoftIron offers a “caddy” that can hold multiple drives that include light indicators for each drive that can be remotely turned on or off depending on the health of the drive.

“The caddy has got some interesting implications, though, because it’s integrated all the way into the stack,” Chadwell says. “With software-defined storage, there’s the concept of rebalancing and the storms that can happen as a result of that. If you pull the wrong drive or you pull an entire caddy’s worth of drives in a software defined storage system, it’ll naturally start to rebalance unless you stop it from doing that and stopping it from doing that usually requires some kind of command line interaction.”

The Ceph Button protects against someone unwitting kicking off that storm by putting the caddy into maintenance mode, pausing the rebalancing. Drives can be swapped out and when the caddy is dropped back into place, the hardware recognizes the act and takes it out of maintenance mode.

Enterprises are going to continue migrating to technologies like Ceph as the amount of data being generated grows and more of it is created outside to datacenters. Garter is predicting that by 2025, 75 percent of data produced will be at the edge and other places other than on-premises. The rise of 5G and growing use of AI will accelerate that trend. More compute and storage capabilities are going to be located closer to where data is being created and applications accessed, so high-performing, highly dense, scalable and open solutions will need to be housed in such areas as manufacturing floors and at the base of 5G towers.

SoftIron has about two dozen customers currently, Moloney tells The Next Platform. Datacenter solutions provider THG is using HyperDrive as the key storage technology for its global THG Ingenuity e-commerce platform. The US Air Force is using HyperDrive as the storage platform for as much as a terabyte of data downloaded from fighter jets after a test or mission that is then analyzed on a local micro-datacenter. Sandia National Laboratory is using SoftIron technology in its Vanguard program as supplemental file and object storage for the lab’s Arm-based Stria HPC cluster, which is part of the larger Astra supercomputer.

Over the next 12 months, SoftIron will use the momentum it’s built and the money collected last year to continue investing in its products, Chadwell says. The plan includes placing additional manufacturing sites around the world. The company says its ability to manufacture and assemble all of its products – including the electronics subassemblies – is a key to what it calls “secure provenance,” which ensures greater data security.

“Our current customer base is already proven that Ceph is viable in the enterprise,” he says. “It’s Ceph plus SoftIron’s unique capabilities make it something that lowers the bar for adopting open-source storage. We intend to continue investing in things like the Storage Manager, additional hardware integrations that ease operations and also improve performance, like implementing solid-state and NVM-Express systems. We also will invest in things that improve its [integration] into existing environments, things like our storage router.”

The work it is doing with the storage router and other technologies will grow the number of use cases SoftIron’s technology can address, which will make the company more relevant to a greater number of organizations, Chadwell says.