Each Open Connect Appliance (OCA) stores a portion of the Netflix catalog, which in general is less than the complete content library for a given region. Popularity changes, new titles that are added to the service, re-encoded movies, and routine software enhancements are all part of the nightly updates, or fill, that each appliance must download to remain current.
By design, OCAs follow a “push fill” methodology. They fill every day during a window of time that corresponds to your off-peak hours. The timing of the fill window is defined in partnership with your network planning team.
The goal is to set the fill window such that:
- It occurs during the trough of your Netflix traffic
- It does not disrupt your inbound traffic peaks
Setting the fill window this way allows the system to continue to download updates and serve clients without negatively impacting playback performance.
Requirements for the fill window, by hardware type:
|Length of time||12 hours||12 hours||12 hours|
|Timeframe||2AM to 2PM (local time) (can be shifted +/- 2 hours)*||2AM to 2PM (local time) (can be shifted +/- 2 hours)*||2AM to 2PM (local time) (can be shifted +/- 2 hours)*|
|Bandwidth||Maximum rate of 1.2 Gbps per appliance||Maximum rate of 0.4 Gbps per appliance||Maximum rate of 0.3 Gbps per appliance|
|Total traffic during fill window||Up to 6.5 TB||Up to 2.1 TB||Up to 1.6 TB|
*All OCAs within a site must begin filling at the same time.
The system might complete filling before the end of the fill window, and it might fill more than once if new content becomes available before the fill window ends.
Most flash-based appliances can fill and serve simultaneously, but some older hardware revisions cannot. These older revisions are configured to fill as much as possible from the storage appliances that are on site, and Netflix traffic is served by the storage appliances on site during their fill period.
For more information about changes to fill patterns that were made after October 2020, see Media-aware content placement
In general, appliances determine where to receive fill using selection criteria that is similar to the steering criteria that is used by Netflix client devices. The following scenarios provide more detail.
If you have specific fill requirements, discuss them with your partner engagement manager during the initial site design process. If you require changes to your appliance’s fill behavior post-deployment, open a ticket.
OCAs will attempt to peer fill from each other in either of these two cases:
- If they are clustered
- If they are on the same subnet
For example, assume that in the following diagram, OCA #1 and OCA #2 are in a cluster. If OCA #1 downloads a movie via a regular cache fill, OCA #2 will then seek to download the movie from OCA #1 instead of via transit or Open Connect peering. Routes to the OCAs must be advertised over Open Connect peering to ensure that the peering session is used for fill traffic.
Peer filling is the most efficient method, however as a second-best option tier filling is considered. OCAs that can see each other’s IP address in their BGP feed (but are not in the same cluster or subnet) will consider filling from each other rather than via a regular cache fill.
There will be cases that cause an urgent need to deploy or update a small subset of titles on the OCAs outside of the normal fill window. This need can arise for various reasons, for example:
- If re-encoding of a popular title is required to repair viewing issues
- To make a new, “day-of-broadcast”, or “day-after-broadcast” title available
When these out-of-cycle fills occur, you will see a small portion of fill traffic for a limited period of time outside of the configured fill window. Doing an out-of-cycle fill instead of waiting until the normal fill window helps to ensure that popular titles can continue to be served from your OCAs (the preferred method) and not via peering or transit. This approach improves the overall efficiency and offload potential of the system. Because out-of-cycle fills are limited to very specific popular titles, they do not negatively impact overall server efficiency - rather, they increase the local offload of popular content.
See also: This blog post about fill