Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

AWS Customer Presenatation - SlingMedia uses AWS

5,801 views

Published on

Brian Lucas, Lead Architect, SlingMedia talks about how they use AWS cloud at the Enterprise Tour - SF - 2010

Published in: Technology
  • Be the first to comment

  • Be the first to like this

AWS Customer Presenatation - SlingMedia uses AWS

  1. 1. Sling Media<br />Hurry up and Wait<br />Hurry up and Wait<br />
  2. 2. Transcoding Challenges<br />Transcoding converts video from one encoded format to another format.<br />Effective real-world transcoding use requires significant CPU power<br />Rendering farms<br />Significant capex<br />Grid environment<br />
  3. 3. In the early days<br />Before Amazon, transcoding large amounts content was limited to an exclusive audience<br />Large companies with budget to support significant capex (rendering farms, entertainment and production houses, SGI)<br />Parallel encoding processes across many machines<br />Machines largely go unused<br />Wasting electricity, IT operations costs, maintenance<br />
  4. 4. Problems with a transcoding system<br />Batch encoding can either be a horizontal scaling effort or a vertical one<br />Faster computers don’t always result in significantly reduced encoding times<br />Some codecs unable to multithread<br />Generally more efficient to do one 1x transcode per CPU (core)<br />Amazon solves the "hurry up and wait" problem in encoding and transcoding<br />
  5. 5. Sling.com At-A-Glance<br />ContentPartners<br />Cloud Ingestion<br />Sling.com<br />Grid<br />Message Queue<br />Encoder<br />Storage<br />Distribution<br />
  6. 6. Encoding Architecture<br />Message Queue<br />SQS<br />EC2<br />Transcode<br />S3 or EBS<br />Storage<br />S3 or Cloudfront<br />Distribution<br />
  7. 7. Encoding: How many servers? <br />Most cost-efficient to maximize 1 hour (~55 minutes) of usage<br />1,000 30 minute videos<br />~15 minutes encoding time per video (2x linear rate)<br />(Total number of videos * average encode time in minutes) / (55 (minutes) * number of cores) = number of servers<br />
  8. 8. Encoding: Non-scientific cost comparison<br />Traditional datacenter<br />$3,000 / server * $1,200/yr electricity (8-core)<br />(1,000 (videos) x 15 (minutes per video)) / ( 55 (minutes) x 8 (cores))= 34 (servers) <br />34 * $3,000 * $1,200 = $142,800 first year capex and run costs<br />EC2<br />(1,000 (videos) x 15 (minutes per video)) / ( 55 (minutes) x 20 (cores))= 14 (servers) <br />14 * $0.68 = $9.52 in costs<br />
  9. 9. Conclusion<br />Elastic grid<br />Ideal for scalable projects, rapid prototyping<br />Collect metrics on CPU, network, disk IO<br />Launch/terminate instances from metrics<br />Encoding/transcoding/map-reduce<br />Cost-savings<br />Terminate instances after certain time period<br />Most cost-efficient to maximize 1 hour of usage<br />Use infrastructure-as-a-service where savings make sense (either cost or operational)<br />

×