18.8 Network latency/bandwidth considerations
NNMi application failover works by exchanging a continuous heartbeat signal between the nodes in the cluster. It uses this same network channel for exchanging other data files, such as the NNMi database, database transaction logs, and other NNMi configuration files. We recommend using a high performance, low latency connection for NNMi application failover when implementing it over a WAN (wide area network).
The NNMi database can become quite large and can grow to 1GB or more even though this file is always compressed. Also, NNMi generates hundreds, or even thousands, of transaction logs during the built-in backup interval (a configuration parameter that defaults to six hours). Each transaction log can be several megabytes, up to a maximum size of 16 MB (these files are also compressed). Example data collected from a Hitachi test environments is shown here:
Number of nodes managed: 15,000 Number of interfaces: 100,000 Time to complete spiral discovery of all expected nodes: 12 hours Size of database: 850MB (compressed) During initial discovery: ~10 transaction logs per minute (peak of ~15/min) ------------------------------------------- 10 TxLogs/minute X 12 hours = 7200 TxLogs @ ~10MB = ~72GB
This is a lot of data to send over the network. If the network between the two nodes is unable to keep up with the bandwidth demands of NNMi application failover, the standby server can fall behind in receiving these database files. This could result in a larger window of potential data loss if the active server fails.
Similarly, if the network between the two nodes has a high latency or poor reliability, this could result in a false loss-of-heartbeat between the nodes. For example, this can happen when the heartbeat signal does not respond in a timely manner, and the standby server assumes that the active server has failed. There are several factors involved in detecting loss-of-heartbeat. NNMi avoids false failover notification as long as the network keeps up with the application failover data transfer needs.
- Organization of this section