In order to make sure queues rarely happen, we have:

• separated batch/import jobs from live/upload jobs
• increased our base capacity to 540 cores
• parallelized upscaling
• implemented analysis of uploads before they enter the encoding queue, so we can launch additional machines ahead of time
• brought down launch times to one minute
• raised our instance ceiling at Amazon, so we can have 1500 heavy machines online (raised over the years, as our requirements grew, from the default of 20).

We are continuously monitoring our system and making improvements to have as little queue time as possible. However, they will still happen from time to time, especially if you exceed your priority job slots or batch job slots.

To learn more about Queues and what you can do to avoid them, read the Queue section in our API docs.