What Makes Us the Faster an the Competition?

One of the first questions prospective customers have about making the switch to cloud media services from traditional on-premise hardware is speed. We've put a lot of effort into ensuring that every step in the typical encoding workflow is optimized for the fastest experience possible. Ingest, queue time, processing, and delivery are all super fast, but don't take our word for it. We put our money where our mouth is with our unique-in-the-industry SLA on queue times.

Encodingspeedtest.com was built by a third party developer and has, through daily tests of both Encoding.com and Zencoder, built up a good set of data over the last year that helps paint a picture of how we have structured our queue algorithm and SLAs. Encodingspeedtest.com submits the exact same job (encoding parameters and source video location of Amazon West) to our platform as well as Zencoder's. You can view the underlying API requests for Encoding.com and Zencoder and give it a try yourself.

How do queue times differ when a cloud provider's infrastructure is busy?

Queue times are defined as the amount of time your job sits in a queue after it has been uploaded to a processing center and before processing begins. Most cloud providers choose to ignore or not report on queue times, but they are an inevitable component of cloud computing, because there is a finite amount of time it takes to spin up a new instance and the OS to boot. In a true elastic capacity queue system, compute resources are automatically scaled in real-time to process incoming requests. A single user can send 100s of jobs in a large batch. For example, if 500 jobs with 10 outputs are sent to a platform within 30 seconds, new instances will need to be started in order to process.

Encoding.com reports queue time as an SLA metric through both the UI and API, and there are two important metrics to consider; average queue time and maximum queue time. Average queue time over the 6 month set of data we looked at was 4 seconds, maximum was 33.37 seconds and the minimum was 2 seconds. Since this test is testing individual rather than batch jobs, it does not accurately test our maximum queue time. However, the concept behind the maximum queue time SLA is that, even in a high volume batch processing situation, you can count a maximum queue time of no more than 4 minutes per job, even if you send 50 or 5000 jobs at one time.

Zencoder chooses not to report on queue times in their job reporting, and it appears they are burying queue times in their download time metric. You can see this illustrated in the download time graph below, where the big spikes in large download time reflect periods when their platform must have been busy and required more time to scale compute capacity.

Accelerated job time when S3 is source location

Encoding.com has developed a unique speed feature which greatly reduces total job time if you are using Amazon S3 as your source for Encoding.com job requests. Encoding.com can directly encode your source from S3 without the need to download the files to an Amazon EC2 instance before processing begins. This can be enabled using the nocopy' command in your API request. You'll notice there are no download times for the jobs run against Encoding.com. Zencoder does not offer this as this option so download time greatly increases total job time.

How do cloud providers maintain consistent processing times?

Encoding.com guarantees the number of dedicated processing cores for your jobs using the turbo (8 core) and twin turbo (16 core) options. We do this to create predictable processing times across all jobs, especially important in an enterprise media workflow with long form or HD content. The mean processing time for the test encoding jobs run on Encoding.com was 81.57 seconds and a median of 80.37 seconds.

Zencoder does not guarantee the number of processing cores for a particular job and instead allocates available compute cores if there system has available spare capacity, if they have less compute resources available when a job request comes in less cores are dedicated to the job resulting in slower processing times. Illustrated in the graph below there is huge variance in the processing time for the same video. The mean processing time for the test encoding job run on Zencoder was 133.78 seconds and a median of 116.85 seconds.

So what does this all mean for our customers? It means that we have paid great attention to every step in our workflow to ensure we are a faster, more consistent, and more predictable encoding experience than the competition.

What are the pain points in your video workflow? Let us know in the comments.

LINK:	http://www.encoding.com/blog/2014/04/07/makes-us-faster-competition/...
	See more stories from encoding

Most recent headlines