Specifications

2-8
Cisco TV CDS 2.5 ISA Software Configuration Guide
OL-24788-01
Chapter 2 Network Design
CDS Workflow
If a VVI is deployed, content is delivered from the Vault Group to the Cache Group in response to
cache-fill calls from the Streamers. The Caching Nodes are explained in more detail in the “Caching
Node Workflow” section on page 2-10.
Within the Streamer array are one or more Stream Groups. The following section describes how the
Stream Groups deliver streams to the subscriber STBs.
Note All servers can be on different subnetworks. However, because of backoffice restrictions, the
externalized IP address is constrained to migrate among servers on the same subnetwork. This means the
Content Store server in an Interactive Services Architecture (ISA) environment can migrate only among
Vaults that are on the same subnet, and the Setup and Control servers can migrate only among Streamers
on the same subnet.
Popularity-Based Caching
Popularity-based caching reduces the write rate to the storage devices on the Streamer and Caching Node
while maintaining the best possible cache-hit rate on the available storage.
To control peak and average write rates to cache (flash or disk storage), the algorithm that determines
when content is written to cache is changed so that only content that is likely to be accessed most often
is cached. Content is only cached if it is more popular than the least popular content that is currently
cached. Otherwise, the content is transmitted from the Vaults to the end-users by way of the cut-through
mode, where content is temporarily stored in the Streamer and Caching Node RAM without ever writing
it to disk or flash storage, and then streamed directly from the Streamer’s RAM to the end-user. When
cache space is needed, the least popular content is evicted from cache first.
The write rate for caching content is determined by the rate at which previously popular content becomes
less popular to the point where it no longer makes sense to keep it in cache, and previously unpopular
content becomes more popular to the point where it does make sense to keep it in cache. Content
popularity is measured by the time-decaying average of the number of play requests on each Global
Object Identifier (GOID).
Previously, all content was written to cache (except when overloaded) and the Least Recently Used
(LRU) content was evicted first.
With the Popularity-Based Caching feature, only popular content is written to cache and the least popular
content is evicted first.
Bandwidth Manager for Thin Pipe
The bandwidth manager controls the traffic leaving the site to any other site and queries all the CDS
servers in the site for the thin pipe mapping configuration of each CDS server. One server in the site is
elected as the bandwidth manager for all servers in the site. A site is defined by the Site Setup page,
which associates groups with a site. Initially, the bandwidth manager allocates bandwidths of whatever
the CDS servers have already committed, provided the committed bandwidths are within the pipe
bandwidth limits; otherwise, the bandwidth manager allocates a percentage of what is committed. After
the initial allocation, the bandwidth manager distributes the bandwidth equally among all the remaining
CDS servers in the site.
Each CDS server in each group reports the bandwidth each one is using to the bandwidth manager every
ten seconds. The bandwidth threshold for each server has an upper limit of 90 percent and a lower limit
of 5 percent. If a server reaches either limit, the server reports this to the bandwidth manager