The week in storage was yet another busy affair that began with news of Samsung’s spacious 15.63 TB SSD, but there is no word on the million-dollar price tag. However, if you buy the limousine of storage, you are guaranteed enough legroom for all of your data, and perhaps a few of its friends.
Not to be outdone, Seagate introduced its speedy NVMe SSDs with a ton of hype but very little substance. In fact, the SSDs are so fast that they outran a name. The details on Seagate’s as-yet unnamed new OCP-compliant SSDs are scarce, but we do know that they are fast, with up to 10 GB/s and 1 million IOPS.
HPE emerged from its successful first quarter of independent operation and announced that it is out to kill data silos. HPE contends that silos are bad (I agree), and they offer the new silo-busting 3PAR StoreServ 20840 as the remedy for your silo-induced pain.
We have not gotten our hands on the aforementioned nameless Seagate OCP-compliant SSDs…but have patience. In the interim, we took the jointly developed Seagate 1200.2 and Micron S600DC SSDs out for a spin. These second-generation 12 Gbps SAS SSDs provide impressive performance and come with an innovative dual-sourcing model.
The OCP (Open Compute Project) Summit is underway, which means a nice spurt of open-source hardware news. Let’s take a peek.
Google Rides With OCP Posse
Google is renowned for its skunkworks projects and top-secret server designs, and I assumed the company would continue that strategy forever due to its ultra-competitive nature. I assumed wrong, and apparently, our Google overlords had a change of heart.
Google announced that it saddled up to ride with the OCP posse, so things are surely about to become more interesting. Google’s first contribution, which it has not actually made yet, is its 48V rack power distribution design. Google has employed the more-efficient 48V design (30 percent more efficient than 12V) for a number of years in its own datacenter, and it roped in Facebook to help develop a common rack specification.
Google indicated that the 48V design is just the beginning of good things to come. The company made vague references to better disk solutions for cloud-based applications, and standardized server and networking management systems as possible areas of future collaboration, then retreated back to its secret lair.
Facebook’s NVMe Lightning Strikes
Facebook is the king of optimized designs, and it modified its existing Open Vault SAS JBOD into a much more powerful Lightning NVMe JBOF (Just a Bunch Of Flash). The new NVMe JBOF provides the PCIe Gen 3 connection directly from the CPU to 60 SSDs.
The new design supports 2.5″, M.2 and 3.5″ SSDs. This enables incredible storage density; if 4 TB 2.5″ SSDs are utilized, a single appliance can provide up to 240 TB of flash-powered storage. Unfortunately, there are architectural limitations that require bifurcating the PCIe connections into two x2 ports, and thermal considerations limit each slot to 14 W of total power. Both of these factors reduce performance, but Facebook’s focus is on storage density (as opposed to wringing the ultimate performance out of every underlying SSD).
The NVMe ecosystem is expanding rapidly, and new features, such as hot-plug, surprise removal support and BMC management, are fostering wider acceptance. The world is embracing flash, but the best way of implementing it is still up for debate. All-Flash Arrays are the most common implementation, but keep your eyes on JBOF designs; they are becoming much more plentiful.
Facebook Drives Wedge Into Networking
What good is storage if you can’t access it? The network is a key enabler of storage performance, and with over 1.5 billion users around the world, Facebook faces extreme networking management challenges inside of its homegrown datacenters.
The company originally contributed to the software-defined networking initiative with its Wedge TOR (Top Of Rack) switch and a set of FBOSS network applications. Facebook announced at the 2016 OCP Summit that it is contributing the second-generation Wedge 100 and its 6-pack modular switch platform to the community.
The Wedge 100 32X100G TOR switch is still in the testing phase and will enter mass production later this year. One of the greatest aspects of the OCP designs is that other manufacturers, especially ODMs, are free to produce any contributed designs. Facebook invited the OCP community to collaborate on the new design and deployment of 100G technologies.
This Week’s Storage Factoid
Facebook claimed that it has saved over $2 billion dollars on its OCP innovations. Stated conversely, Facebook has eviscerated $2 billion in sales from the OEMs’ balance sheets. The real savings total for all of the members of the OCP posse, which includes heavyweights such as Apple, Microsoft, Bank of America and Goldman Sachs (the list goes on), is probably immeasurable.
The OCP movement is an amazing amalgamation of companies working together to serve a common good, and even though the large hyperscalers are the main beneficiaries of the technology, the advantages trickle down to all of us. The OCP movement is forcing the incumbent OEMs to become more competitive as they fight the rise of whitebox/ODM vendors, and this equates to savings and innovation for the broader audience.
The en masse migration to the cloud is underway, but as with all emerging technologies, there will be a number of speed bumps. On-prem datacenters will never disappear, but the continuing development and extension of the OCP ecosystem will help the participating cloud vendors drive pricing down even further.