Interpreting the sysfs statistics

An example of a statistics report and the meaning of the values.

seconds_active

The attribute seconds_active is a single value attribute (see Table 1) and simply gives the seconds the FCP device has been active.

requests

The attribute requests is a three-valued attribute that provides the number of requests that are processed since FCP device activation split into the areas of (in that order):

Input
Output
Control

The following example shows that three input, ten output, and five control requests were issued since FCP device activation:

[root]# cat /sys/class/scsi_host/host0/requests
3 10 5
[root]#

megabytes

The attribute megabytes is a two-valued attribute that provides the number of megabytes transferred in and out. The following example shows that 3 MB were received and 6 MB were sent out since FCP device activation:

[root@T6360007 host0]# cat /sys/class/scsi_host/host0/megabytes
3 6
[root@T6360007 host0]#

utilization

The attribute utilization is a three-valued attribute, showing the utilization of the processor, bus, and FCP channel over the last two seconds. The FCP channel continuously refreshes the values covering the utilization of the individual sections over the past two seconds. These values cannot be reset manually.

cmd_latency, read_latency, and write_latency

Each latency provides seven values as follows:

value, minimum fabric latencies [microseconds]
value, maximum fabric latencies [microseconds]
value, summarized fabric latencies [microseconds]
value, minimum channel latencies [microseconds]
value, maximum channel latencies [microseconds]
value, summarized channel latencies [microseconds]
value, amount of requests

No interpretation or modification of the values is done by the zfcp device driver. The individual values are summed up during normal operation of the FCP device. An overrun of the variables is neither detected nor treated. You must read the latency twice to make a meaningful statement, because only the difference between the values of the two reads can be used.

Example: After reading the file twice, you have the following values:

Type	1st read	2nd read	Difference
Fabric	403	821	418
Channel	115	163	48
Count	21	23	2

The average fabric latency (see Figure 1) over two readings is

418/2 = 209 microseconds

The results for the other values can be calculated accordingly.

Figure 1. SCSI latency breakdown

The graphic describes the constituent parts of SCSI latency.