As my inaugural blog post on this forum, I thought it would be worth establishing a baseline on key storage technologies. As a former IT director turned HP Storage strategist, I know that storage can often be a cryptic technology domain. My goal here is to channel that inner IT guy. I will do my best to translate storage topics into business value, discuss emerging trends and answer questions you may have on how storage is changing in a world of virtualization, cloud and big data.
Here begins the first in a three-part series on the changing storage landscape.
Midway through a recent 11-day trip visiting with a few hundred of our storage customers in Malaysia, India and Singapore, I began thinking about the changing world of storage choices.
For example, the other day I found myself sitting in an airport using as many as three devices—smartphone, tablet and laptop. In trying to recharge my mobile ecosystem—uploading travel photos to Snapfish, sending status updates on Facebook, downloading emails remotely from Microsoft Exchange, running a query to see what my international data roaming charges were looking like and streaming some HD video to pass the time—I realized that I was using virtually every conceivable storage choice. DAS, NAS, SAN and object storage protocols were all in play. It’s also likely that the back-end disks I was hitting were connected to the world via iSCSI, Fibre Channel and SAS running both spinning media and solid state drives.
It also struck me that five years ago, this massive amount of airport storage use would’ve made me stand out like the geek I am, but on this trip I was surrounded by many other people doing the exact same set of activities.
The world has certainly changed, and we can see this not just in the rate of data growth and storage consumption, but also in the types of storage being consumed. With this as a backdrop, here is the first of a series of blog postings offering an overview of the alphabet soup of ingredients in the storage landscape along with a perspective on how they are evolving.
First, Direct Attached Storage (DAS)
Direct Attached Storage. The name is pretty self-explanatory. A disk subsystem that is directly connected to a host rather than going through a switched network, thereby giving the host exclusive access to the disks. The category obviously includes disks internal to a physical server, but in the storage realm we most often think about JBOD (“just a bunch of disks”) shelves attached to a server via an SAS cable.
Traditionally, JBODs were added to servers to either expand capacity or to deliver higher IOPS through additional spindle count. Modern operating system and application vendors have recently been getting smarter about storage—turning application performance and building storage logic into the application layer that was previously the realm of the disk array manufacturers. This move has shifted how DAS is positioned and what use cases it is applicable in.
For example, with Microsoft Exchange 2010, the folks in Redmond released the ability to have multiple Exchange Servers, each with DAS replicating to each other through what they call Database Availability Groups (DAGs). Most companies had Exchange traditionally connected to SAN and lots of disks to provide both resiliency and IO. With Exchange 2010, Microsoft improved the IO profile and, with DAGs, is actively positioning DAS as a connectivity option to get large, low-cost mailboxes—without sacrificing performance or resiliency.
Why today’s DAS is also more nuanced than the JBODs of old
“Shared DAS” or “Direct Area Networks” (DAN?) are variations on a theme whereby multiple hosts can be connected to a set of array controllers, bypassing the need for a switched infrastructure. On the small side, we have modular arrays that can support 4 servers directly attached in a resilient formation. A nice little array deployment plan for a midmarket cluster at a lot less cost than building up a SAN infrastructure. On the high-end, we have tier-1 disk arrays with massive amounts of host connectivity and emerging converged networking infrastructure that can also bypass multi-layer switches. An example of the latter is the Virtual Connect FlexFabric model within the HP BladeSystem. Using “Flat SAN technology,” you can direct attach more than 700 servers to a 3PAR disk array without the need for any SAN switches or host HBAs.
Another evolution of DAS is the emergence of independent public cloud providers looking at deploying massive farms of storage servers and low-cost JBODs as an alternative to large dedicated disk arrays. Not to be confused with those deploying traditional applications in IT-as-a-Service (ITaaS) paradigms, these independent public cloud providers are bringing completely new applications to market (think Dropbox and the like) and, in order to reach competitive price points, they often pursue home-grown or open-source storage software as a route to pool low-cost, high-capacity disks together in unique ways.
The DAS space will be interesting to watch as application vendors continue to build more logic into applications and as traditional array vendors fight for share. This trend benefits the systems vendors that have a strong affiliation with compute infrastructure and DAS, but hurts traditional “storage only” vendors who have made their bread and butter on shared storage infrastructure.
In the next week or so, I’ll move on down the alphabet with a discussion on NAS, SAN, unified and object storage.