Skip to main content

Storage Performance In Entertainment And Content Creation

Transcoding Multiple Streams (Intel Quick Sync)

Overall StatisticsTranscoding, Quick Sync
Elapsed Time03:54
Read Operations144 885
Write Operations9233
Data Read16.00 GB
Data Written756.30 MB
Disk Busy Time41.72 s
Average Data Rate410.91 MB/s

According to Cisco
, global Internet video traffic surpassed global peer-to-peer (P2P) traffic in 2010. And, by 2012, Internet video will account for over 50 percent of consumer Internet traffic. Amazingly, that number does not include the amount of video exchanged through P2P file sharing.

This underscores the pervasiveness of streaming video, which went into Intel's decision to dedicate silicon on its Sandy Bridge-based processors to accelerating video encode and decode. Dubbed Quick Sync, the company makes it possible to transcode an entire unprotected Blu-ray movie in under 20 minutes. We already took a very thorough look at output quality and found Quick Sync to be surprisingly good. For more information, read Video Transcoding Examined: AMD, Intel, And Nvidia In-Depth.

That's exactly why we profiled transcoding in our previous examination of storage in office productivity. Unfortunately, that was somewhat limited in scope because it only involved transcoding a single video clip. Using the same applications, it's often possible to transcode multiple video clips simultaneously. Fraps is a perfect example. The software truncates each video file to 4 GB, so we have a total of six video clips from our Battlefield 3 gameplay.

While CyberLink's MediaEspresso only allows a total of four simultaneous transcoding operations, that still counts as encoding multiple files at the same time. Overall, the access pattern is still sequential involving 128 KB transfers. Queue depth is where we see things diverge. With more files accessed at the same time, the majority of the operations are queued between two and five deep.

I/O Trends:

  • 16% of all operations occur at a queue depth of one
  • 47% of all operations occur at queue depths between two and four
  • 30% of all operations occur at queue depths between five and eight
  • 82% of all data transferred is sequential
  • 88% of all operations are 128 KB in transfer size