Solid State Disk (SSD) is the medium that storage managers are looking
at to address storage performance problems in their environment.
Deduplication is what they look at to address out of control capacity
requirements. Rarely do you see the two technologies linked, but these two
may be a perfect match.
Assuming that you do have a real storage I/O problem, SSD will usually
eliminate it. In fact, for many storage problems, it's overkill, but SSD
is becoming more cost effective than stringing together a group of
mechanical hard drives in a RAID group. What is needed is something in
between. In other words, a slightly lower performing SSD that is also
less expensive could be an ideal solution for most storage I/O problems.
While many SSD vendors are exploring the use of higher density MLC
based flash for this purpose, another option is to use deduplication.
high performance deduplication options available to vendors that can
be implemented with little or no performance loss. While the risk of
creating noticeable I/O performance loss on SSD is certainly greater
than on mechanical drives, as stated earlier, SSD may have some I/O to
spare. For that potential trade-off in performance, you get added
capacity. Take for example a 500GB PCIe SSD card: these currently go for less than $15,000. Add deduplication with a modest effectiveness
of 5X storage efficiency, and that 500GB card may now be able to hold
2.5TBs. Consider the same combination in a server or desktop
virtualization environment, and you may be able to take that utilization
rate up to 10X or more. For many environments 5TBs is the entire
Ironically because of the I/O write issues with Flash SSD, inline
deduplication may be the right way to implement deduplication with SSD.
While the assumption would be that inline deduplication would add
significant latency to primary storage SSD, if used in the right
environment it may speed things up. As we discuss in our article on
one of their biggest roles is managing writes. Inline deduplication, in
the right environments, could reduce significantly the number of writes
being sent to the devices, reducing the amount of write logic that has
to be processed by the flash controller.
There are several products currently shipping that perform
deduplication on SSD technology, and with the announcement of several
primary storage deduplication API sets, more should be on the way. The
combination of significantly increased SSD capacity for a small
reduction in performance may be a winner.