Hi Ajit,
Zookeeper
Zookeeper utilizes a leader election mechanism. The minimum
installation (beside all-in-one/single node) of Private Cloud will use 3
Zookeeper processes. After 3, additional Zookeepers are be added due to high
availability requirements in multi-region (multi-DC) installations.
We recommend colocating Zookeeper and Cassandra so from than
point of view, most customer have the same number of Zookeeper than Cassandra.
Zookeeper is only used for configuration and management. It
doesn't plays a role during runtime traffic.
Cassandra
Routers, Message Processors and Cassandra are the three
components on the critical path. Meaning, they actively participate on runtime
API traffic.
Edge uses replication factor 3 for Cassandra, so 3 nodes are
required on the minimum installation (beside all-in-one/single node).
Additional Cassandra nodes can be added based on requirements. When expanding
the cluster we normally double its size, meaning if you have 3 nodes you will
add 3 more. This allows you to keep a balanced ring.
What requirements play a role in Cassandra capacity planning?
Among others:
- TPS: How much traffic is your installation sized for.
- Policies type: Edge provides two family of policies. A)
policies that don’t need to store any data. i.e. JSON-XML and b) policies that
generate and/or retrieve data from Cassandra. i.e. Oauth. The number of
policies (B) used in your proxies drives Message Processor-Cassandra
interaction.
- Storage: Cassandra is designed to scale horizontally.
Depending on your storage needs, the way to add storage capacity will be adding
new nodes to the ring.
- Availability: Cassandra availability for local quorum operations
is calculated as N/2 + 1. Where N is the number of nodes in a region. This
means, in a 3 nodes installation, you can lose 1 node without impacting availability.
Message Processor
Similar to Cassandra capacity planning comment above, TPS, Proxy
complexity (number and type of policies), availability and other factors drive
the number of MPs.
In order to determine how many you need evaluate
requirements such as:
- Current and expected (next 18 months) TPS
- Average TPS
- Peak TPS
- Peaks duration and/or characteristics of any seasonality
they may experience or expect in their API traffic.
- Availability and resiliency
- Expected number of physical installations (Apigee planets)
- Are planets single DC, multi-DC?
- If multi-DC, how many?
- If multi-DC, are they going to run API traffic active/active
or active/passive?
- If one DC fails, what is the expectations for the others?,
should a single DC be able to handle 100% of the API traffic?
- SSL termination. Where it will happen?
- Network zoning requirements.
- Estimated average payload size
- Proxy complexity
Please let me know if you need further assistance doing capacity planning. We will be happy to help.
I also encourage you to attend Apigee Private Cloud Operational Training, this hands-on training covers all the items above and a lot more.
LinkedIn
Twitter
