I've had a number of opportunities to speak with Rob Commins over the years, usually at trade shows where he is proudly showing off Tegile Systems' feature rich storage arrays. And each time I speak with Rob, I walk away with learning something new in the storage world.
This interview was no exception. In this Q&A discussion, Rob Commins, vice president of marketing at Tegile Systems, talks about hyrbid storage technologies, the difference between hybrid storage and an all-flash memory storage array, caching and tiering benefits, and more.
VMblog: What features are the most important when evaluating a Hybrid Storage vendor?
Rob Commins: There are big differences in how
vendors implement what is being called hybrid storage technologies. Here are a
few key questions to consider when stacking up alternatives:
- Is the vendor using less reliable
consumer grade flash drives?
- Are data reduction
technologies in-line or post-process?
- Can I access the array with
both block and file interfaces?
- What data management features
such as snapshots, replication, thin provisioning are
included?
VMblog: What are the main differences
between a new cache optimized hybrid array and a traditional array with flash
added to it?
Commins: There are significant
differences that cache optimized designs offer that traditional arrays simply
can't deliver. The most significant is the ability to run data reduction
technologies such as deduplication and compression in line before data is
cached. This has drastic effects on the efficiency of the array's cache.
Let's
say dedupe and compression are delivering a 5:1 reduction ratio. With in-line reduction, the effective size of 2TB of physical cache is actually 10TB! That
means that five times more data can be cached and cache hit ratios will get a
huge boost. Traditional arrays with post-process data reduction only get
capacity efficiencies in the hard disk pools. This is great, but grossly misses
the mark of a truly cache optimized hybrid array.
VMblog: What are the acquisition and
operational cost differences of hybrid storage and traditional storage?
Commins: The cost advantages of a cache
optimized hybrid array are huge and can be realized instantly. When users are
buying up to 80% less capacity, there are immediate acquisition cost
implications that are easy to measure. You don't need an ROI calculator to
ascertain the benefits of buying 20TB versus 100TB of capacity.
Operationally, our customers see
similar cost reductions. One of our customers sent me a note that he was amazed
to replace 115U of legacy storage with 12U of Tegile's gear. He was seeing the
same performance and capacity while reducing space/power/cooling by 89%.
Amazing.
VMblog: What is the benefit of implementing a Hybrid Solution rather than all-flash memory storage array?
Commins: There are places where an
all-flash array fits better than hybrid, but typically, hybrid arrays fit the
bill at a much lower cost and with a considerably easier to manage footprint.
The key question between all-flash and hybrid is "how often can I withstand a
latency hit above 1-2 ms?" If you are running a massive OLTP database for a
credit card company, an all-flash array makes sense. If you have a medium sized
enterprise running databases, consolidating VMs and looking to deploy VDI, there
is no comparison: hybrid is the way to go.
I mentioned manageability. With an
all-flash array, you still need a place to put capacity centric data - typically
unstructured files. That means buying a separate array that is optimized for
$/GB versus $/IOP. With a hybrid array, you have a single well balanced system
that can resolve the age-old imbalance between capacity and performance. You
don't have to play Tetris with your data.
VMblog: Why are caching and tiering capabilities important?
Commins: Some people think caching and
tiering are the same. They're not. In a cache optimized hybrid array, caching
is a real time, in-line process that keeps hot data in DRAM and SSD and lets
cool data migrate down to spinning disk. It is always happening. Tiering is
very different - it uses a process that at a fixed time interval (usually once
or twice a day), the array's data management software will sweep for what I like
to call IO density. "Where are the hot regions of data to move to fast media,
and where is the cold data to move to slower HDD?" With the agility of business
applications and new use cases like VDI, users can't wait 12 or 24 hours for
their data to end up in the appropriate media. Caching is a far more agile and
efficient means to deliver performance and efficiency.
VMblog: How does separating the metadata
from the primary data path help me?
Commins: Metadata (data about the data)
is a really big deal in storage systems. Metadata is used to manage extremely
important functions such as RAID, snapshot pointers, and deduplication tables.
This metadata is typically interleaved with user data to keep some locality
between the two. There is an inherent problem with that. Metadata IO needs to
be extremely fast in a cache-optimized array. If metadata is sitting on
spinning disk, it will inherently be accessed slowly. An array that separates
metadata and only stores it in fast media such as DRAM and SSD (well protected,
albeit), the array can run all of the metadata operations at extremely fast
speeds, making every application run faster.
##
Once again, special thanks to Rob Commins for taking time out to speak with VMblog.
About Rob Commins
Rob Commins has been instrumental in the success of some of the storage
industry's most interesting companies over the past twenty years. As Vice
President of Marketing at Tegile, he leads the company's marketing strategy; go
to market and demand generation activities, as well as competitive analysis.
Rob comes to Tegile from HP/3PAR, where he led the product marketing team through
several product launches and 3X customer growth over three quarters. Rob also
managed much of the functional marketing and operations integration after
Hewlett Packard acquired 3PAR. At Pillar Data Systems, he was at the forefront
of converged NAS/SAN storage systems and application-aware QoS in mid-range
storage. Rob is also a veteran of StorageWay, one of the first storage services
providers that launched cloud services.