I would like to ask for advice about Slow operations (network - other) metric. Documentation describes this metric as: "The number of slow operations caused by other factors than latency or loss rate, which is one of the detailed reasons in network category as calculated using the primary reason for slowness algorithm. Note that this includes only sucessful operations. Failures and aborted operations are not taken into account." I understand this definition but I can't imagine what could be the other factors.
Please, can you provide me any detailed information about this metric and "the other factors"?
Solved! Go to Solution.
Please look at this documentation page for DCRUM 12.3:
To sumarize, a page is marked Slow operations (network - other) when it goes through the following analysis :
-1. the operation is slow (response time higher thant the performance threshold)
-2. then, the page is slow because of the response download time ( time for the page to be sent from the server to the client)
-3. then, the page is considered as a slow operation with
Throughput responsible for operation time degradation
-4. then, the page is considered as a slow operation not because of the Round Trip Time, Loss Rate or number of Hits.
I hope this help you.
thanks for your replay. I understand described analysis but I can't imagine another reason for slow operation than RTT, Loss Rate or Number of Hits in point 4. Could you tell me what can be another reason?
Jakub, algorithm is looking for dominant fault domain. It may happen that no domain (RTT, LR) is dominant (or, more precisely, there's no dominant domain AMD is recognizing), it may also happen that other low-level network-related factors are responsible for long transmission time (throttled window size, TCP slow start..., chattiness or some others ) so then algorithm is reporting this cause as "network - other".
The name „Network – other” might be a bit misleading as it is actually “Network – multiply reasons”.
It applies to a situation in which the operation was slow, the network time took more than 50% of the operation time, and none of the following conditions were true:
• Percentage of retransmissions above threshold
• Percentage of request time above threshold
• Percentage of response time above threshold
• RTT (latency) above threshold
• Percentage of redirect time above threshold
In such case the AMD is not able to clearly identify one single cause of slowness and sets the cause to “Network – multiply”.
All thresholds above are configurable on the AMD: per decode and per software service.