alphaspirit - Fotolia
I have been testing some product alternatives in the server-side or software-defined storage space, comparing them to solutions promoted by leading hypervisor vendors, including VMware and Microsoft. On this journey, I recently met up with StarWind Software, an outfit based in the Boston area, with development in Kiev, Ukraine. They're also the company that arguably invented virtual SAN technology (though they failed to trademark the term), so they deserve some serious consideration.
The first question that needs to be addressed is why a virtualization administrator would prefer to move outside the comfort zone of a "one throat to choke" relationship with his/her preferred hypervisor vendor to consider a virtual SAN product from a third party. The thought of buying a "pre-integrated" hardware/software stack from a single source has enormous appeal -- at least to anyone too young to remember life in an IBM data center circa 1980.
I am that old, however. In my first data center job, the entire IT hardware/software stack was dominated by Big Blue and you deviated from its prescribed architecture at your own risk. IBM had become a de facto standard and everyone had to comply with its rules for plugging and playing with the IBM stack if they wanted any purchase whatsoever in an IBM-dominated world.
That lock-in contributed a lot of great technology, but it also helped to make IT an extremely expensive component of the business. Over time, those costs set the stage first for a flirtation with IT outsourcing (service bureau computing) during the Reagan recession, and ultimately for the distributed computing "revolution" of the 1990s.
Today, hypervisor vendors seem to be taking a page from the old "one vendor is best" playbook to make the case for a new single-vendor model for the data center: hypervisor-controlled computing. In the worst-case scenario, we run the risk of locking ourselves into another master/slave relationship where we're the ones chained to the oars. At best, we just make the operational side of our data center worse than it already is.
If, and this is quite likely, we end up using multiple hypervisors in our data center (in addition to having some critical apps that aren't virtualized at all), we're going to end up with multiple silos of data behind different hypervisor software-defined storage (SDS) strategies along with some "legacy storage." VMware pretty much locks up access to its storage with its Virtual SAN, excluding all non-VMware servers. Microsoft at least enables SMB access to the storage it controls provided it's configured as a scale-out file server.
Hypervisor SAN gotchas
What if you could simply deploy a third-party SDS that supported all applications, virtualized or not, with their data? That's what a few companies, including StarWind Software, claim they can do. They provide block and file access to their virtual SAN storage to all comers. Sounds pretty good.
Moreover, when you dig down into VMware, you quickly learn that its SDS freezes out a lot of smaller firms and even some large ones. For the big guys, VMware offers no path to an all-silicon data center. You scale by adding disk to nodes and nodes to clusters. Flash can be used in the mix, but it isn't included in scale-out capacity. That will irritate some large IT shops that see an all-silicon future.
Meanwhile, smaller firms are likely to be put off by VMware's storage node requirements. First, you need a minimum of three nodes with all their storage maintained in an identical configuration, even as things scale. To start, you're looking at a combined hardware and software licensing cost of between $30K and $40K per node. That's quite a bite out of a small shop's IT budget. It might even be the entire budget for some modest environments.
So, for small and large firms, the VMware one-stop shop may already seem too expensive or limited. And from a technical perspective, many architects are put off by the lousy way VMware SDS uses flash from a write perspective.
Depending on your virtual machine stack, you'll be hammering your flash memory cache with small writes, which is to say you may burn out expensive flash devices more quickly than you thought. An alternative is to coalesce your writes, and stack them up in DRAM until you can write them efficiently to flash in fewer but longer content write operations. This functionality already exists in StarWind Software's product. VMware advises us to wait another year or so.
Microsoft, by the way, has some limitations that may be off-putting to architects with a strategic view. Like VMware, Microsoft's SDS approach, called Clustered Storage Spaces, isn't terribly friendly to flash when used as a write cache. In fact, the vendor's deduplication increases the number of writes since data is first written as is and then subsequently reduced by the dedupe algorithm. For those who want dedupe technology, the inline capability in the StarWind product is arguably more robust.
Another thing Microsoft shares with VMware is its penchant for nodal equipment definitions that might make the infrastructure too pricey for a smaller firm. This starts with the requirement that each node have external SAS JBODs (Microsoft uses some of the SAS technology to lock files and volumes, which is one of the points VMware uses to argue why it is not truly software-defined), which are quite a bit pricier than their SATA cousins. StarWind supports both, plus PCI Express flash devices.
As you can see, hypervisor vendors are building out their SDS solutions in a manner that both addresses the architectural requirements of some of their customers -- perhaps the majority -- and favors the vendor's concept of how SDS should work. An SDS-only vendor is in a better position to provide complementary support for hypervisor operations while helping users to (1) avoid getting locked into a particular vendor's concept and architecture, and (2) realize an integrated storage environment that will support a number of workload types and their storage requirements.
StarWind Software, by the way, is only one potential solution to the issues raised above, but it has earned bragging rights for being the first provider of virtual SANs long before VMware or Microsoft seized on the idea.
About the author:
Jon William Toigo is a 30-year IT veteran, CEO and managing principal of Toigo Partners International, and chairman of the Data Management Institute.
- CW500: A roadmap to software-defined everything – Paddy Power Betfair –ComputerWeekly.com
- CW ANZ: Taming the data beast –ComputerWeekly.com
- Success Story: Com Hem and Scality RING –Scality
- The Business Value of HPE Object Storage with Scality RING for Data-intensive ... –Scality