To diagnose a network element, right-click and select View Metrics. You can troubleshoot these network elements:

  • Tier – The performance chart (top left) shows the rate or application performance outliers on the relevant nodes (Errors and Slow/Very Slow/Stalled Calls) and the Key Performance Indicators for all connections used by those nodes (Errors and PIE). 
  • Network Link – The performance chart (top left) shows the rate of Performance Impacting Events for all member connections of that link.  
  • Network Connection – You can troubleshoot a connection from the Connection Explorer or from the Connections tab in a link popup.
  • You can troubleshoot a transaction from a Transaction Snapshot. 

    • You can right-click an application flow and select View Network Metrics.
    • You can also drill down to a node where a transaction delay/stall/error occurred and access the Network tab.

To troubleshoot a tier, link, or connection in the Network Dashboard, right-click the network element and select View Metrics. The top-left chart in the dashboard shows the overall network or network/application performance of the element. To troubleshoot the element, search for correlations between the performance chart (top left) and the other charts on the page. 


To best highlight metric spikes and variations, you can switch between the linear and logarithmic scale in each chart. Click the settings button (top-right corner of the chart) to switch between scales.

root-cause correlation charts

Right-Click Chart Descriptions


Default Monitoring Mode

Network Impact on Transactions (Tiers)

This chart highlights the possible impact of Performance Impacting Events on the app-transaction outliers (Transaction Errors and Slow, Very Slow, and Stalled Calls shown in the Transaction Scorecard).

If you see spikes in transaction outliers and correlated spikes in PIE or errors, this indicates that the network is affecting application performance. Search for correlated spikes in the other charts to identify specific issues and root causes: connection errors, packet loss, re-transmissions, high-latency connections, and so on.


Host Stack KPIs

Interface errors indicate issues on the node's physical interfaces. See Node Metrics in Network Dashboard and Metric Browser.

TCP Wait sockets can result in significant delays or errors for the application, or service, that relies on that socket. Many simultaneous WAIT sockets can prevent applications and services from creating new connections.


Network PIE - Contributors

Use Performance Impacting Elements (PIE) to identify the location of actual or potential bottlenecks:


Network Errors - Contributors

Use this chart to identify network errors that can affect application performance:

  • FIN Errors – Errors when tearing down a TCP connection
  • Syn Blackholes - Connection attempts that went unanswered and resulted in a failure
  • Syn Resets – Connection attempts that were explicitly refused by the other host
  • RST on Established - TCP connection resets


Traffic throughput for the application of interest on the network element. This chart measures application payload data only; TCP, IP, and other packet headers are excluded.KPI

Connection Rate Info

Spikes in this chart, that coincide with spikes in slows calls or applications errors, indicate a possible issue with how the application is using TCP.

  • New Connections – Number of new connection attempts
  • Connection Errors – The rate of TCP connection setup (SYN) errors and connection teardown (FIN) errors
  • RSTs (resets) – A TCP reset is an immediate closing of a connection. Not all connection resets indicate a problem, but you should investigate any spike in resets that coincides with a spike in application errors or slow transactions/calls. Connection resets can occur for various reasons, such as:
    • Inability to create connections by the server
    • Intermediate network elements, such as load balancers, firewalls, and so on, due to misconfiguration or other errors
  • Current Established Connections

TCP Loss (mille)

The number of packets lost (sent but not received) per 1000 packets sent. "Per mille" is a percentage with one additional digit of precision. TCP detects lost packets and retransmits each lost packet until it receives an ACK (acknowledgment) from the peer. Spikes in TCP Loss indicate that the network is over utilized.KPI

Retransmissions per Minute

  • Data Retransmits – The percentage of packets that were retransmitted. This metric includes SYN and FIN retransmissions
  • SACK Retransmits – The percentage of data packets that were retransmitted due to selective acknowledgments (SACK, a TCP Feature)

Latency (RTT) Comparison

This chart compares the average TCP round-trip times (RTTs) for different types of packet request and responses.

  • Handshake RTT – Average RTT for the initial 3-way handshake (SYN, SYN/ACK, ACK) to set up a connection. If Handshake RTTs are significantly higher than Initial RTTs, this indicates that the client node is taking a long time to set up the connection.
  • Initial RTT – Average RTT for the initial SYN packets (between SYN and SYN-ACK or SYN-ACK and ACK). If Initial RTTs are high, the delay may be due to an intervening firewall, asymmetric routing, or other network issues that causes the network to treat SYN packets differently from data packets.
  • RTT – Average TCP RTT for data request/responses—Time for TCP Data segments to be transmitted and acknowledged from the peer. This is different from Application Response Time metrics that measure the time an application takes to process a request. 

Connection Lifetime

TCP is most efficient when using long, stable connections. Connection setups and teardowns are very time-consuming and resource-intensive. The more short-term connections are generated, the worse TCP performance will be because most of the time is spent creating connections and because "Slow-Start" in TCP optimal TCP bandwidth is not achieved.