Manual Chapter : Collecting and Viewing TCP Statistics

Applies To:

Show Versions Show Versions

BIG-IP Analytics

  • 14.1.3, 14.1.2, 14.1.0
Manual Chapter

Collecting and Viewing TCP Statistics

Overview: Viewing TCP statistics

You can set up the BIG-IP® system to gather information about TCP flows to better understand what is happening on your networks. The system can collect TCP statistics locally, remotely, or both. You can view these statistics in graphical charts, and use the information for troubleshooting and improving network performance.
The statistic reports for both TCP and FastL4 show details about RTT (round trip time), goodput, connections, and packets. For TCP, you can also view statistics for delay analysis. You can save the reports or email them to others.

Task Summary

Creating a TCP Analytics profile

Before you can create a TCP profile, you must have provisioned the Application Visibility and Reporting (AVR) module.
A TCP Analytics profile directs the system to store TCP statistics about specific entities for use in diagnosing network problems. The Application Visibility and Reporting (AVR) module includes a default TCP Analytics profile called
tcp-analytics
. You can edit the values in the default profile, or create a new one, as described here.
  1. On the Main tab, click
    Local Traffic
    Profiles
    Analytics
    TCP Analytics
    .
    If
    Analytics
    is not listed, you need to provision Application Visibility and Reporting (AVR) first.
    The
    TCP Analytics
    screen opens.
  2. Click
    Create
    .
    The New TCP Analytics Profile screen opens, inheriting values from the system-supplied TCP Analytics profile.
  3. For
    Profile Name
    , type a name for the profile.
  4. From the
    Parent Profile
    list, select the profile from which you want to inherit settings.
    The default profile is often used as the parent profile.
    The new profile inherits the values from the parent profile. If the parent is changed, the inherited values in the new profile also change.
  5. To make the fields editable, click
    Custom
    in the upper right corner.
    If you don't need to change any of the values, just use the default profile instead.
  6. For the
    Statistics Logging Type
    setting, verify that
    Internal
    is selected. If it is not, select it.
    Selecting
    Internal
    causes the system to store statistics locally, and you can view the charts on the system by starting at the Main tab, and clicking
    Statistics
    Analytics
    .
  7. For
    Statistics Collection
    , leave the default,
    Client side
    , selected.
    This option specifies where the system gets the statistics from.
  8. In the Associated Virtual Servers area, specify the virtual servers that use this TCP Analytics profile to capture TCP statistics:
    1. For the
      Virtual Servers
      setting, click
      Add
      .
    2. From the Select Virtual Server popup screen that displays, select the virtual servers to include, and then click
      Done
      .
      Only virtual servers previously configured to use TCP protocol or FastL4 (Type Performance Layer 4) display in the list (because the data being collected applies to TCP or FastL4 traffic). Also, you can assign only one TCP Analytics or HTTP Analytics profile to a virtual server; therefore, the list displays only virtual servers that have not been assigned either of these profiles.
      The system attaches the profile to the virtual servers you added.
  9. In the Statistics Gathering Configuration area, for
    Collected Entities
    , select the entities for which you want the system to collect information.
    The more entities you enable, the greater the impact on system performance.
    Entity
    Description
    Virtual Server
    Saves statistics about the virtual server. Note that this is always selected to start with.
    Remote Host IP Address
    Saves the IP addresses with which traffic was exchanged.
    Remote Host Subnet
    Saves the addresses of the subnets with which traffic was exchanged.
    Next Hop Ethernet Address
    Saves the addresses to which traffic is being routed.
    Continent
    Saves the name of the continent with which traffic was exchanged.
    Country
    Saves the name of the country with which traffic was exchanged.
    Region
    Saves the name of the region with which traffic was exchanged.
    City
    Saves the name of the city with which traffic was exchanged.
    Postcode
    Saves the name of the postal code with which traffic was exchanged.
  10. Click
    Finished
    .
The system creates the TCP Analytics profile. If the BIG-IP system is exchanging traffic with clients, TCP statistics are collected for the virtual servers and collected entities specified in this profile.
To view RTT statistics for Layer 4 traffic managed using a FastL4 profile, you also need to enable
RTT from Client
and
RTT from Server
(as appropriate) in the FastL4 profile.
If later you decide you want to store TCP analytics remotely, you can use the external Statistics Logging Type and specify a remote publisher to specify where to send the statistics.

Viewing TCP statistics

Before you can view TCP statistics, you must have created a TCP Analytics profile that is logging statistics internally on the BIG-IP system. The TCP Analytics profile also needs to be associated with one or more virtual servers.
You can view TCP statistics in the Analytics charts.
  1. On the Main tab, click
    Statistics
    Analytics
    TCP
    .
    The RTT statistics screen opens.
  2. For
    Time Period
    , you can adjust the time frame for which to display the data.
  3. To look at the statistics from a different point of view, for
    View By
    , select the category of data to display in the chart.
    You can also click an item in the Details list to drill down and display more specific statistics.
    The screen displays data in the categories for which you are saving statistics in the TCP Analytics profile.
  4. Click any item on the menu bar to see different TCP Analytics charts.
    Click This
    To View These Statistics
    RTT
    Round trip times from the BIG-IP system to the remote host and back.
    Goodput
    Throughput at the application level used to review overall network performance. It shows total throughput aggregated for all connections on the configured entities.
    Delay State
    The aggregate time spent in each delay state by all connections. This is only available for connections with a TCP profile, not FastL4.
    Connections
    New and closed connections. It also shows mean connection length, measured from when Analytics starts collecting data (which may be from a mid-connection iRule) to when it stops.
    Packets
    Packets sent, packets received, and packets lost.
    The system displays the different charts, and you can adjust the time period and view by settings on all the charts.
  5. To save the charts to a PDF or to email the chart, click
    Export
    and specify the option to use.
    To use email, the BIG-IP system requires an SMTP server which you can configure at
    System
    Configuration
    Device
    SMTP
    .
The TCP statistics are available to use for evaluating network performance. You can save the reports to track the differences in performance over time.

Sample TCP RTT statistics

This figure is a sample TCP statistics chart showing round trip times (RTT), or how long it takes for outgoing TCP packets on the client side to be answered by the server. When you hover over the chart, it shows the RTT minimum, RTT maximum, RTT average (mean), and the RTTVAR mean values. You can use these statistics to help gauge application performance.
To view RTT statistics for Layer 4 traffic managed using a FastL4 profile, you need to enable
RTT from Client
and
RTT from Server
(as appropriate) in the FastL4 profile.
Sample TCP RTT statistics chart
Sample TCP RTT chart

Sample TCP goodput statistics

This figure is a sample TCP statistics report showing goodput sent and received values from the client side. Goodput shows throughput at the application level over a period of time. You can use these statistics to understand network performance.
Sample TCP Goodput statistics chart
Sample TCP Goodput

Sample TCP delay state statistics

This sample TCP statistics report shows the causes of delay states. Here the primary delay state cause is application latency either on the client, or server side.
Sample TCP Delay State statistics chart
Sample TCP Delay State Chart
The delay states, described in the following table, are color coded in the chart. You can hover over the part of the chart you are interested in to display the delay states and their values. These states apply to outgoing data. Analytics picks the first listed state that matches the current situation.
State
Description and What to Do
3WHS
3-way handshake that starts a TCP connection. Analytics will accrue time in this state only if it can estimate the round-trip-time of the SYN or SYN-ACK that it sent.
RETX
Retransmission. TCP is resending data and/or waiting for acknowledgment of those retransmissions. This may indicate lossy links in the data path, or overly aggressive congestion control (for example, a profile with
Slow Start
disabled or improperly set
Packet Loss Ignore
settings). Activating rate-pace in the TCP profile may also help.
CLOSING
The BIG-IP® system has received acknowledgment of all data, sent the FIN, and is awaiting acknowledgement of the FIN. If the FIN goes out with the last chunk of data, you might not see this state at all. If there is a major issue on the client side, the issue may be that the servers are configured for
keepalive
(to not send FIN with their last data).
WAITFORACK
The BIG-IP system has sent all available data and is awaiting an ACK. If this state is prevalent, it could be a short connection, or possibly either the upper layers or the server are forcing TCP to frequently pause to accept new data.
APP
The BIG-IP system has successfully delivered all available data. There is a delay either at the client, the server, or in the layers above TCP on the BIG-IP system.
RWND
Receive-window limited. The remote host’s flow-control is forcing the BIG-IP system to idle.
SNDBUF
The local send buffer settings limit the data in flight below the observed bandwidth/delay product. Correctable by increasing the
Send Buffer
size in the TCP profile.
CWND
Congestion-window limited. The TCP congestion window is holding available data. This is usually a legitimate response to the bandwidth-delay product and congestion on the packet path. In some cases, it might be a poor response to non-congestion packet loss (fixable using the
Packet Loss Ignore
profile options) or inaccurate data in the congestion metrics cache (addressable by disabling
Congestion Metrics Cache
, the ROUTE::clear iRule, or the
tmsh
command
delete net cmetrics dest-addr <addr>
).
NAGLE
TCP is holding sub-MSS size packets due to Nagle’s algorithm. If the NAGLE state shows up frequently, disable
Nagle's algorithm
in the TCP profile.
RATEPACE
TCP is delaying transmission of packets due to rate pacing. This has no impact on achievable throughput, and no action is required.

Sample TCP connection statistics

This sample TCP connection report shows the average connection length in milliseconds, and the number of connections opened and closed during the last hour. If new connections are outpacing closed ones, that means the system may be unsustainably loaded.
Sample TCP Connections statistics chart
Sample TCP Connections Chart
You can change the information that is displayed in the chart and the Details table by changing the
View By
setting. For example, you can view by
Countries + Regions
to see where the connections are originating.

Sample TCP packets statistics

This sample TCP packets report shows the number of packets lost, sent, and received during the last hour. Packet loss is typically caused by network congestion, and can impact application performance. In this example there are is no packet loss.
Sample TCP Packets statistics chart
Sample TCP Packets Chart
You can drill down into the statistics. For example, on systems with multiple virtual servers, applications, or subnet addresses, you can investigate specific entities that might be having trouble. If users are having difficulties with an application, from the
View By
list, select
Applications
. In the Detail list, click the application to zoom in on the statistics for that application only.

Sample iRule for TCP Analytics

You can create a TCP Analytics profile that uses an iRule to collect the statistics. In the profile, for
Statistics Collection
, do not select either
Client Side
or
Server Side
. Let the iRule handle it.
For example:
# start collection for one subnet only. when CLIENT_ACCEPTED { if [IP::addr [IP::client_addr]/8 equals 10.0.0.0] { TCP::analytics enable } } when HTTP_REQUEST { # must check subnet again to avoid starting for all # connections if [IP::addr [IP::client_addr]/8 equals 10.0.0.0] { # make stats queryable by URI TCP::analytics key "[HTTP::uri]" } }
For more information about iRules®, refer to
devcentral.f5.com
.