Follow

InfiniBox measures the latency that is associate with write and read requests as follows:

  • Total latency - the total time the I/O operation takes, from the time of arrival of the SCSI command to the InfiniBox to the time the final response was sent to the host. The total latency may increase due to host or fabric delays, QoS-added latency, synchronous replication or internal latency
  • Internal latency - the time InfiniBox spends handling the I/O operation (i.e. without waiting for more data from / to the host). It is affected by system resource use but is neither affected by host and fabric delays nor by QoS added latency

Consider the following examples:

Example 1: write request

The order of events is:

  • t0 - the host initiates a write request
  • t1 - InfiniBox receives the write request, performs write-related operations and returns a response to the host
  • t2 to t5 - InfiniBox may further communicate with the host in order to accept the write request (for example, a write request of the size of 128kb is sent to InfiniBox in 64kb chunks)
  • t6 - the host receives a final acknowledgement from InfiniBox 

On the image below:

  • internal latency
  • total latency

Example 2: read request

In the following example, InfiniBox receives a read request of 128KB. The order of events is:

  • t- the host initiates a read request 
  • t1 - InfiniBox receives the read request, parallel-fetches the data in two chunks of 64KB, and sends the first chunk
  • t2 - the host informs InfiniBox that it is ready for the next data chunk
  • t3 - InfiniBox sends the second chunk 
  • t4 - the host receives all of the data that was requested

What is monitored by InfiniMetrics

As of InfiniMetrics 4.2, we present graphs for both local total latency and local internal latency:

  • SAN total latency by I/O type - the total time the I/O operation takes, broken down by I/O operations
  • SAN internal latency vs. total latency  - the difference between internal and local latency only for volumes that are not synchronously replicated


Was this article helpful?
0 out of 0 found this helpful

0 out of 0 found this helpful

Comments