Distributed Storage Done Right
Debate revolves around scalability and cost effectiveness
- By Mahesh Saptharishi, Ph.D.
- Dec 01, 2012
Scalable system architectures
have been the source of many
debates within the security industry.
A particular point of the
debate revolves around distributed
video storage and its ability to scale
quickly, easily and cost effectively.
As distributed storage continues to gain broad mainstream
appeal, it’s worth revisiting the discussion.
Architectural Advantage
Modern networks are designed to scale with applications
that follow a few-to-many streaming model—a
small set of servers stream to many clients. Multicast
is one example of such a network feature. But traditional
IP video surveillance is inherently a many-to-few
streaming application—many cameras stream to a few servers.
Thus, IP video surveillance networks are plagued by bottlenecks
not anticipated by today’s network infrastructure. By distributing
primary storage and processing to the edge and allowing end
users to intelligently stream on demand, distributed storage and
processing systems adapt to the strengths of today’s network
infrastructure rather than being limited by its constraints.
Tests have shown that distributed systems eliminate more
than 90 percent of the network traffic of a traditional, centralized
surveillance system because only the video that is being actively
viewed is transmitted across a network—no video needs
to be backhauled to central storage. If centralized archiving of
important content is needed, then the backhaul can be scheduled
to accommodate the load on the network. Additionally,
customers can choose lower resolution and bitrate streams for
live viewing while still recording full-resolution, high-quality
video at the camera, further reducing bandwidth consumption
without sacrificing the forensic value of evidence.
Finally, distributed systems eliminate the need to manage
and archive video in a large, common repository—the greatest
computing and storage resource bottleneck in a central architecture.
The result is a system that is network and computer efficient,
allowing systems to scale from a single camera to thousands
of cameras with minimal infrastructure.
Market Adoption
There has been a significant increase in companies integrating
with distributed storage providers over the past year. The mainstream
support and acceptance of distributed storage by large
VMS companies will ensure that the architectural benefits of
distributed storage will be available to mainstream end users
whether they require one camera or thousands.
Important Considerations When
Evaluating Long-term Storage Options
- Solid-State Disk (SSD) storage is an expensive solution for
long-term storage. SSDs are thought of as an appealing option
for long-term storage due to their perceived reliability
and data throughputs. But not all flash storage technologies
are equal in performance. Those that measure up to expectations
are cost prohibitive for long-term storage applications.
- HDD alone is not reliable. HDD storage is the current standard
for long-term storage because it provides the best cost
per gigabyte with theoretically sufficient reliability to meet
customers’ needs. Spinning magnetic media storage technology
has evolved significantly during the past decade and
a half. However, as any experienced integrator or end user
knows, hard drive reliability is a constant concern and an expensive
source of frustration.
Unfortunately, HDD failures are simply a fact of life. Accounting
for failure concerns often requires elements of redundancy
and fault tolerance to be factored into the system
architecture. Depending on the nature of the installation, more
complex fault-tolerant storage architecture increases up-front
costs and installation time. In large-camera-count installations,
the cost of building in fault tolerance is a small percentage
of the overall system costs and thus may not influence the
storage decision.
However, in smaller installations where cost is a significant
factor and decision driver, the end user who does not build a
redundant storage system might be forced to suffer through
downtime and lost video.
- Hybrid storage delivers high reliability at a compelling price
point. As a general rule of thumb, the lifetime of an HDD is
correlated with the number of spinning hours of the disk. In
most user-driven applications, disk access is often “bursty”
and not continuous. In other words, even with heavy workloads,
the expectation is that the disk is not spinning 24/7.
Typical surveillance applications require continuous 24/7
disk access, accelerating most vendor mean time between
failures (MTBF) estimates.
A widely cited 2007 study by Carnegie Mellon University’s
Bianca Schroeder and Garth Gibson showed that even enterprise
HDDs don’t offer significantly longer MTBFs for continuous
access applications. The duo’s work also brought into
question some advantages of RAID 5.
There are hybrid storage solutions available that combine the
robustness of industrial SLC flash storage with the cost advantage
of HDDs, up to 750GB. The combination works by first
filling the flash storage with video. Once the flash storage is full,
it spins up the HDD storage, writes the stored video to the HDD
and then suspends the HDD.
More specifically, the capacity of the solid-state storage is
sized such that the HDD is spinning for a small fraction of every
hour. Because the HDD is suspended and not spinning for the
majority of the hours in a day, the disk usage model is consistent
with what manufacturers account for in their design. Thus,
in video surveillance applications, the MTBF can be four times
longer than using an HDD alone. Additionally, advanced analytics
control compression quality based on video content, reducing
the total volume of video both for continuous recording
and event recording. This further reduces HDD spin time and
contributes to increased MTBF.
Important Considerations When Evaluating Short-term Storage Options
There are numerous situations when an integrator, remote
monitoring company or end user will need short-term storage,
typically a few days to a week of storage. In some cases this is
because the cameras are being continuously monitored and any
evidence of loss or a crime is found within minutes or hours, let
alone days.
In other cases, the surveillance solution includes long-term central
storage, using distributed storage mainly as a fault-tolerance
mechanism protecting against network failures and downtime.
Many cameras and encoders offer SD cards as a solid-state solution
that provides reasonable storage capacities at moderate prices.
There are better solid-state storage options than consumer
SD cards. Flash memory actually comes in several variations
differing in speed, reliability and cost. SLC NAND flash memory
is the high end of the flash variants, providing the most reliable
storage available with a high cost to match. Each memory
cell in SLC NAND has 10 times longer lifetime than its cheaper
sibling, MLC NAND.
MLC NAND flash fills out the middle and lower end of
flash storage options with multiple variations within the MLC
category. MLC NAND flash in the mid-range provides a good
combination of speed and reliability without breaking the bank.
The larger the capacity of flash storage, the fewer times a single
memory cell is rewritten. Thus, theoretically, MLC flash storage
with 10 times the capacity of SLC flash will have about the
same reliability for sequential-write applications (such as storing
video). Unfortunately, when MLC flash storage capacities
go beyond a point, read/write latencies and errors also increase.
Thus, MLC NAND flash is a good solution for applications
requiring short-term storage. MLC NAND-based SSDs up to
256GB in capacity are now fairly common and have been shown
to have good reliability.
The lowest tier of MLC NAND flash is used in most consumer
SD cards. These types of low-quality flash storage are not
designed for applications that require gigabytes of data per hour
to be written 24 hours/day. Though an SD card slot is a common
choice on many surveillance cameras and encoders, reliability
will be a serious issue if the SD card experiences continuous
video writes. If installers or end users are not careful about their
choice of consumer or professional SD cards (those that use
higher-end MLC NAND), premature media failure is inevitable.
Edge storage creates scalable surveillance networks. As discussed,
one of the most reliable choices is a hybrid-storage option
that delivers high reliability at a compelling price point.
When considering long- or short-term storage options, make
sure you understand the different storage choices, what they deliver
in terms of reliability and capacity, and when you should
make trade-offs between cost and reliability.
This article originally appeared in the December 2012 issue of Security Today.