After years of same-old same-old, we are heading into a period of innovation and change in the IT industry. New software approaches are altering the way systems are assembled and are affecting hardware configurations at all levels down to the components.
The primary driving force in all of this is the increase in system-level horsepower that’s resulted from SSD/flash storage and much faster networks. Total system performance has closed the gap on CPUs and caught up with Moore’s Law. In the future, we’ll see system performance doubling every two years, rather than just CPU power.
The new horsepower is challenging the industry’s creative powers. Obvious evolutions include faster and larger SSDs, but 2015 will also bring new classes of drives, with specifications characterized around wear-out. For instance, there will be “archiving” SSDs, which, with a write-once mentality driving them, will use 3D TLC structures to increase capacity. Next year, we’ll see flash products up to 5 TB. By the end of 2016, these may match the largest available hard drives, at 10 TB. This will build a debate over SSD-only datacenters.
At the performance-end of the SSD market, we’ll have plenty of choice in Non-Volatile Memory express-based drives (NVMe). These are screamingly fast units, rated to 1 million random IOPS or beyond, that are well-matched to in-memory databases. SATA Express looks like next year’s choice of interface for these drives, with motherboards in the pipeline already.
In-memory databases present their own challenges to conventional system design. Systems with as much as 6 TB of DRAM are available, though the sweet spot seems to be at 1 TB in 1U or 2U server. SanDisk has a companion to DRAM which puts flash in a DIMM and takes advantage of the memory bus speed. This will gain momentum in 2015, with instant-on capabilities and increased capacity per flash DIMM. 3D stacked cell technology in flash will be the driver for this.
In 2015, we'll also see the containerization trend have a growing impact on storage installations. Storage performance is already under pressure for virtualized systems, and using containers increases that. One alternative is to distribute storage in a "Virtual SAN,” with the drives co-resident with the server engines. With a careful distribution of data, which may prove too difficult, this should radically speed up instances. 2015 will be the year VSANs rise or fall as a solution.
Another trend will be “naked” drives, with just Ethernet interfaces. Seagate and HGST are already delivering product, and in both cases, the fit with “hot” storage software stacks like Ceph has come together well. At the same time, Ceph is proving to be a very flexible open-source storage solution, allowing for more traditional appliance designs, too. It has universal storage support (block-IO, object, and file access) and can run with its elements structured anywhere from virtual instances to dedicated appliances.
With Red Hat now supporting Ceph, a mainstream open source stack will lead to creation of a low-cost storage appliance market most likely driven by the Chinese ODMs who serve Google, AWS, and Azure in huge volumes. In 2015, these storage white boxes will begin to take noticeable marketshare from the traditional vendors, which are trying to reposition themselves as software and services providers.
Storage packaging will see subtle changes in 2015, as more buyers opt for pre-assembled clusters at either the rack or the container level. This approach will allow most enterprises the opportunity to consider Open Compute Project types of storage, thereby getting some of the minimalist benefits the big cloud providers enjoy. The ODMs are stepping up to the plate on this, offering all the needed components as a “Lego” kit, as well as integration services.