Edge Storage


John Honovich opened a discussion on this topic, along with a lot of good analysis at his web site:

http://ipvideomarket.info/report/sd_cards_video_surveillance_storage

This plays right into my most recent post on intelligent storage in the camera. The webinar I held last week had record attendance, so there seems to be interest in this subject.

I’ve been watching this shift in storage technology for the last few years, and I went through a lot of the same math that John spells out so well. I’d like to add a few added twists to the topic that show why I’ve come to a different conclusion.

The most important comparison that John points out is the cost of a 1 TB hard drive versus an 8 GB or 16 GB SD card. When you look at the cost in $/GB, the SD card is 21X more expensive. So, even though flash memory is dropping in price faster than hard drive memory, how could it ever catch up in the next five years?

Here’s an important piece missing from this comparison: You can’t just plug in your hard drive into the network, you need a storage server. When you get done adding the storage server, which includes all the hardware needed to manage lots of streams from many video cameras onto the hard drives, then the cost per TB is much, much higher. The typical end user price I’m hearing today is around $2,000. There is a wide range in pricing and options here, but this is the average I’ve heard.

That’s 20X more than the price of the 1 TB hard drive. Since the cost of the SD card is the total cost needed, this wipes out most of the cost difference with using SD cards compared to hard drives in a storage server.

If you put the hard drive in the camera directly, as we do in our iCVR, then you don’t need all of that server overhead. That’s why I think that today, putting a hard drive in the camera, ends up being the lowest cost storage solution.

For the next big factor that changes the math, take a look at this whitepaper by Uptime Institute:

http://www.uptimeinstitute.org/wp_pdf/%28TUI3008%29Moore%27sLawWP_080107.pdf

This shows a growing problem with the costs of data centers. It has become a lot more expensive to support servers these days than it used to be. Since the latest servers are far more dense in storage capability, and far more powerful, they also draw a lot more electricity to run, and they generate a lot more heat, which requires more electricity to cool the rooms down and higher support costs. On top of this, the numbers of servers in data centers are growing fast.

Take a look at Table 1 on page 3. It shows there that for every $1,000 you spend today on servers, there is a total cost of $3,510 over three years. That means you now need to pay the cost of your server over again each year, for overhead, power, and replacement of hard drives as they fail, in today’s data center. This is based on conservative electricity costs. In many parts of the world, it is going to be higher.

This includes the costs of hard drives failures, and that’s another big factor here, since flash memory lasts a lot longer. In fact, storage failure in cameras should become rare, while hard drives in DVRs and servers fail after 3-5 years.

While this overhead, electricity, cooling, support and hard drive failure costs aren’t usually calculated into the cost of servers when video surveillance systems are bid, that should change. IT managers realize this is a growing issue and that cost needs to be added into the true total cost of ownership when comparing.

These issues of power consumption and data center overhead, plus failure rates, is one of the reasons that solid state storage is expected to make serious inroads into replacing hard drives in storage data centers. The cost of the overhead and electricity and support has become bigger than the cost of the servers. Solid state storage uses far less power, has longer life expectancy, and can run at higher temperatures reliably.

The use of solid state drives is just now starting to be used in data centers, mainly for special applications, since cost is still higher. But most projections I’ve seen say that in 5 years, this is going to change and SSD will start replacing hard drives.

We can expect to see the same thing happening in laptops, since lower power consumption means longer battery life, and with the rugged abuse that laptops get, solid state storage makes a lot more sense from a reliability standpoint.

In five years, the growth rate of hard drives should start slowing significantly, as solid state memory starts taking over.

These are some big factors that clearly push the equation in the favor of SD cards as they drop in price, and they get large enough to provide 30 days of storage. They aren’t quite there yet.

The latest reports I’ve seen say that 64 GB SDXC cards will be available early next year, and 128 GB cards will be introduced by the end of 2010. If this is true, we could see 1 TB cards being introduced in 5 years, although I think it might take a year or two longer than that, since the rate of advancement will probably slow down a bit as the demand for such large storage cards will not be as high.

I agree completely with a lot of the other points that John raises: That video management software can’t deal with storage in the camera today. This needs to be addressed. Also, the storage on most SD cards in cameras cannot be accessed like you would a DVR (Mobotix is the only exception I know of, besides our own cameras with built-in storage). In fact, some cards need to be removed physically to access the stored video. So, this is a long way from ready for most cameras.

However, the cost impacts above suggest to me that SD cards in IP cameras (or microSD cards) will become the standard in five years. The largest megapixel cameras may require hard drives, but most of the cameras will have plenty of storage capability.

This is of course just a guess, and the thing about technology predictions like this is that you have to always wonder what factors you overlooked. Will a new technology be introduced that changes everything? Or does someone solve the problems with hard drives and server overhead costs?

Storage at the edge of the network, especially right in the camera, continues to generate interest, as I mentioned in my last blog post.

However, not many realize all of the reasons why it is so compelling.

For example, the whole idea for centralized recording of video comes from the traditional data center. It is natural to think that what makes sense for data centers is also the right solution for video storage.

It turns out, however, that this is not the best architecture for IP surveillance, which is why the industry has run into so many problems with bandwidth, storage costs, system reliability, etc.

Storing video in DVRs from analog cameras is actually closer to the edge of the network than IP cameras with server storage. IP cameras have actually been going the wrong direction. The answer that makes everything far simpler, and overcomes many of the biggest problems, is to go the other way and move storage and intelligence further to the edge, right into the camera itself.

It makes more sense to treat cameras as sensors, not just data sources, because sensing is what they are really about. This becomes much more obvious once you include video analytics in the cameras. Then it becomes much easier to see why sensor systems work better with distributed storage and intelligence. Where do you want to do the sensing? Back at a central location or right there at the sensor? Why send 100% of the video back, when you only access 1% of it? Why waste 100X as much bandwidth as you need?

I just finished a whitepaper that describes why intelligent storage in the camera is going to have a big impact on IP video systems. It’s a look at the future of where technology is going and why.

You can download the whitepaper here:

http://www.videoiq.com/products/resource-center/whitepapers/

I will also be holding a free webinar on the subject in a few weeks. Here is a link to register, if you are interested:

https://event.on24.com/eventRegistration/EventLobbyServlet?target=registration.jsp&eventid=153908&sessionid=1&key=BAB80E0D27E9D00175CACB0907737599&partnerref=july8&sourcepage=register

The leading video management systems companies are beginning to recognize the importance of intelligent storage in the camera, but I think it is going to end up being a much bigger trend than they realize. There are a number of future shifts in technology that are going to make it even more compelling. For example, solid state storage, megapixel cameras and the move to video web services.

Some have suggested that as bandwidth gets easier to find and storage gets cheaper that these will make centralized storage and intelligence more popular, but this is all based on thinking of cameras as if they were data sources, not sensors. Yes, there are lots of PC applications and data that run much more efficiently in the data center, but sensor systems work much better at the edge. Everyone has simply been looking at this the wrong way.

Video intelligence offers a lot more than just detection. It changes the whole system for the better.