The Challenge for Next-Generation Enterprise Video Platforms
At least one column in any given table is designated as the primary key. This column needs to contain unique information for every single field in that column and establishes the relationship between various tables and database records. Without a unique primary key, there’s a risk that content in a given record will be mistakenly served up when an SQL query or report is run.
For streaming content, a record could vary from something as simple as a DASH segment or as complex as an entire alternate audio stream.
Content that doesn’t fit in to a given table or column, such as additional metadata about an existing record that doesn’t yet have a column in the table, potentially could be abandoned. In other words, data types added later—or used infrequently—aren’t a good fit for the field-based relational database model
In the past 5 years, though, another kind of database has appeared, driven partially by big data concept and “dirty data” that doesn’t fit well into the relational model. These databases go by the terminology of NoSQL, although the name’s a bit of a misnomer since content can still be queried.
Two primary types of databases using the NoSQL model are document-based and map-reduce indexes. We won’t spend an inordinate amount of time on these, but let’s take a look at each of them briefly, and their applicability to various types of video platforms.
Document-Based NoSQL Databases
These databases use a JavaScript Object Notation (JSON) as a way to mark up a large document for later partial retrieval. Document-based databases rely on the premise that all the information about a given user should reside in a single document, rather than be spread across multiple tables as it would be in an RDBS. The idea might be to store all the videos a single user has consumed in one document, along with all pertinent contact information, so that the IT and training departments can easily find that information.
Document-centered databases work well when a few users have consumed large amounts of content, but the majority of users have not. An example of this would be Users A and C each viewing 100 videos per week, while Users B and D only consume 10 videos per week. In a traditional RDBS, there could potentially be a table for every single video, plus a table that contains users, while a document-based NoSQL database would only have one document per user.
Map-Reduce Indexes
The second type of NoSQL database creates a map of potentially pertinent information, then reduces that information in a series of preindexing steps. The reason for this mapping and reduction, followed by indexing of those key store values, is to speed up database queries.
In some ways the map-reduce concept is similar to the streaming media concept of “pre-fetching” content, where the first 5–10 seconds of multiple video clips are downloaded to the client’s player ahead of time. In the case of streaming media, the idea is to allow rapid viewing of the beginning of any video a user chooses. From a database standpoint, the idea is to limit the amount of content the database query needs to search through, which in turn makes responses faster.
A map-reduce index might include not only the most popular enterprise video content but to also HTTP Live Streaming (HLS) or DASH segments for the most popular bandwidth and resolution combinations.
The other big benefit of modern database structures is something called sharding. Sharding is the process of spanning multiple servers in a database cluster, or even multiple clusters, in a way that balances the load of hundreds or thousands of simultaneous database queries.
Sharding not only allows an EVP solution to move beyond the legacy limited single server/ database structure, but also to create opportunity for solutions that need to take advantage of shared resources in the cloud
Hybrid vs. Fully Cloud-Based Solutions
That leads us to the last point in our exploration of next-generation enterprise video platforms: Hybrid solutions might be the best approach, but we’re still in a state of flux.
There’s little doubt that some things—such as simple transcodes, or even single-channel encoding—can be done better on premises. This use of resources is often referred to as “on prem,” following the concept of telecom installations late in the last century, where phone lines were pulled into an enterprise location, connected to an expensive “middleware” box known as a PBX, which then routed calls to whomever the PBX had been programmed to route a given number to.
Along the way, companies began to look at the idea of having some services in the cloud, although they weren’t called cloud services at the time. Many companies went to a Centrex model, which essentially virtualized the primary services of a PBX, such as call-forwarding, allowing for hundreds of handsets to be added to an enterprise location without requiring the company to buy a PBX capable of handling all those handsets.
The next move was toward VoIP, where the PBX was further virtualized, essentially using the data switch and router to create a software-only telephony solution. Every handset was nothing more than a data terminal, and any handset from an enterprise location could suddenly be moved around and plugged into a different Ethernet port without requiring someone to come in and program the PBX to a particular phone line.
Why the history lesson on enterprise telephones? Because we’re moving in the same direction with streaming in an enterprise environment.
The end points in this case, though, should be thought of as both receiving and sending devices. In some instances, such as the smartphone or tablet, the same device can create content and consume different content. In other instances, probably for the majority of employees, only content consumption matters.
What’s curious about the move within the EVP ecosystem, though, is the tendency of EVP providers to suggest skipping the middle step and going to fully cloud-based solutions.
“We consider cloud-native architecture a next-generation solution,” one EVP vendor says. This cloud-only approach should be fully assessed, as it often doesn’t account for content created and consumed in the same location, instead forcing all content to the cloud first.
A better model, and one that enterprise will probably swing toward in late 2015, is the hybrid approach. The routing of content will be virtualized, as perhaps will the storage, but the use of on-premise appliances to generate content will regain popularity.
On the consumption front, the swing toward soft clients and apps may also see some moderation, with a new breed of set-top boxes emerging to fill the gap. We’ll look at both of those options in the 2015 Streaming Media Industry Sourcebook’s “State of the Enterprise” article coming in March.
One thing is clear as we enter 2015: Enterprise video platforms and delivery solutions are moving rapidly to catch up, in terms of complexity and even capacity requirements, with the traditional entertainment-focused media delivery platform. Scale on an enterprise solution, especially for some multi-national corporations, could easily meet or exceed that of an entertainment platform, and we suspect enterprise will solve the issue of multiple-device/ multiple-screen delivery at least a year before general entertainment solutions emerge.
This article appears in the January/February 2015 issue of Streaming Media magazine as "Next-Generation Enterprise Video Platforms."
Related Articles
As graduates accustomed to the "flipped classroom" take the place of retiring professionals, enterprises must be prepared to deliver knowledge to them via video anywhere, anytime, and on any device
29 Dec 2014
Employees in a variety of departments are viewing and creating work-related video, and they're doing so on both desktops and mobile devices.
04 Dec 2014
Within enterprise video departments, such as Symantec's SymantecTV, understanding viewers' needs is crucial for getting good results.
26 Aug 2014