TCP Traffic - VM Test
Since most popular applications rely on the TCP protocol for their proper functioning, traffic monitoring at the TCP protocol layer can provide good indicators of the performance seen by the applications that use TCP. The most critical metric at the TCP protocol layer is the percentage of retransmissions. Since TCP uses an exponential back-off algorithm for its retransmissions, any retransmission of packets over the network (due to network congestion, noise, data link errors, etc.) can have a significant impact on the throughput seen by applications that use TCP. This test monitors the TCP protocol traffic to and from a guest, and particularly monitors retransmissions.
Target of the test : A Hyper-V / Hyper-V VDI server
Agent executing the test : An internal agent
Output of the test : For a Hyper-V server, one set of results will be reported for every powered-on VM on the server.
For a Hyper-V VDI server, one set of results will be reported for the user who is currently logged into each virtual desktop on the server
|
Measurement | Description | Measurement Unit | Interpretation |
---|---|---|---|
Segments received by VM |
Indicates the rate at which segments are received by the guest. |
Segments/Sec |
|
Segments sent by VM |
Indicates the rate at which segments are sent to clients |
Segments/Sec |
|
Retransmits by VM |
Indicates the rate at which segments are being retransmitted by the guest |
Segments/Sec |
|
Retransmit ratio from VM |
Indicates the ratio of the rate of data retransmissions to the rate of data being sent by the guest |
Percent |
Ideally, the retransmission ratio should be low (< 5%). Most often retransmissions at the TCP layer have significant impact on application performance. Very often a large number of retransmissions are caused by a congested network link, bottlenecks at a router causing buffer/queue overflows, or by lousy network links due to poor physical layer characteristics (e.g., low signal to noise ratio). By tracking the percentage of retransmissions at a guest, an administrator can quickly be alerted to problem situations in the network link(s) to the guest that may be impacting the service performance. |