profiler

Profiler intermediate plugin

The ipfix-profiler-inter plugin is an intermediate plugin for IPFIXcol (IPFIX collector). It profiles IPFIX data records and fills in metadata information according to given set of profiles and their channels.

Introduction to profiling

The goal of flow profiling is multi-label classification (based on a set of rules) into user-defined groups. These labels can be used for further flow processing. The basic terminology includes profiles and channels.

A profile is a view that represents a subset of data records received by a collector. Consequently, this allows surfacing only the records that a user needs to see. Each profile contains one or more channels, where each channel is represented by a filter and sources of flow records. If a flow satisfies a condition of any channel of a profile, then the flow will be labeled with the profile and the channel. Any flow can have as many labels as possible. In other words, it can belong to multiple channels/profiles at the same time.

For example, let us consider that you store all flows and besides you want to store only flows related to email communications (POP3, IMAP, SMTP). To do this, we can create a profile "emails" with channels "pop3", "imap" and "smtp". When a flow with POP3 communication (port 110 or 995) is classified, it will meet the condition of the "pop3" channel and will be labeled as the flow that belongs to the profile "emails" and the channel "pop3".

Example of a profile hierarchy:

live (channels: ch1, ch2)
├── emails (channels: pop3, smtp, imap)
└── buildings (channels: net1, net2, net3)
    ├── office (channels: http, voip)
    └── factory (channels: http, voip)

The profiles can be nested and create a tree hierarchy (as shown above). A flow source of a channel in a profile can be only one or more channels of the direct parent of the profile to which the channel belongs. For example, the channel "http" in the profile "office" can use only "net1", "net2" or "net3" channels as sources. The exception is the highest level i.e. "live" profile. This profile must be always present, has exactly this name, and its channels will receive all flow records intended for profiling.

How does flow profiling work? In a nutshell, if a record satisfies a filter of a channel, it will be labeled with the channel and the profile to which the channel belongs and will be also sent for evaluation to all subscribers of the channel. For example, let us consider the tree hierarchy above. All flow records will be always sent to all channels of "live" profile as mentioned earlier. If a flow record satisfies the filter of the channel "ch1", the record will be labeled with the profile "live" and the channel "ch1". Because the flow belongs to the channel "ch1" it will be also sent for evaluation to all subscribers of this channel i.e. to the channels of the profiles "emails" and "buildings" that have this channel ("ch1") in their source list. If the record doesn't satisfy the filter, it will not be distributed to the subscribers. However, if the record satisfies the channel "ch2" and the channels of the profiles "emails" and "buildings" are also subscribers of the channel "ch2", the record will be sent them too. Thus, the record can get to any channel in different ways but labeled can be only once.

For now, following plugins support profiling provided by this plugin:

• lnfstore - Convert and store IPFIX records into NfDump files. For more information see manual page of the plugin.
• profilestats - Create and update RRD statistics per profile and channel. For more information see manual page of the plugin.

Plugin configuration

The collector must be configured to use Profiler plugin in the startup.xml configuration. The profiler plugin must be placed into the intermediate section before any other plugins that use profiling results. Otherwise, no profiling information will be available for these plugins.

<collectingProcess>
    ...
    <profiles>/path/to/profiles.xml</profiles>
</collectingProcess>
...
<intermediatePlugins>
    ...
    <profiler>
    </profiler>
    ...
</intermediatePlugins>

The plugin does not accept any parameters, but the Collecting process must define parameter <profiles> with an absolute path to the profiling configuration.

Structure of the profiling configuration

This section describes parts of the XML configuration file. Example configuration for an email subprofile can be found in a section below.

Keep in mind that profile and channel names must match a format that corresponds to the variables in the C language. In other words, a name can have letters (both uppercase and lowercase), digits and underscore only. The first letter of a name should be either a letter or an underscore. Consequently, the names must match regular expression [a-zA-Z_][a-zA-Z0-9_]*.

Warning: Using complicated filters and multiple nested profiles have a significant impact on the throughput of the collector!

Profile (<profile>)

<profile name="...">
    <type>...</type>
    <directory>...</directory>
    <channelList>...</channelList>
    <subprofileList>...</subprofileList>
</profile>

Each profile has a name attribute (i.e. <name>) for identification among other profiles. The attribute must be unique only among a group of profiles those belong to the common parent profile. In the definition of each profile must be exactly one definition of following elements:

• <type>
  Profile type ("normal" or "shadow"). Normal profile means that IPFIXcol plugins should store all valuable data (usually flow records and metadata). On the other hand, for shadow profiles, the plugins should store only metadata. For example: In case of the lnfstore plugin, only flows of normal profiles are stored. Others are ignored.
• <directory>
  The absolute path to a directory. All data records and metadata that belong to the profile and its channels will be stored here. The directory MUST be unique for each profile!
• <channelList>
  List of one or more channels (see the section below about channels)

Optionally, each profile can contain up to one definition of an element:

• <subprofileList>
  List of subprofiles that belongs to the profile. The list can be empty.

Channel <channel>

<channel name="...">
    <sourceList>
        ...
        <source>...</source>
        ...
    </sourceList>
    <filter>...</filter>
</channel>

Each channel has a name attribute (i.e. <name>) for unique identification amongst other channels within a profile. Each channel must have exactly one definition of:

• <sourceList>
  List of flow sources. In this case, a source of records should not be confused with an IPFIX/NetFlow exporter. The source is basically a channel from a parent profile from which this channel will receive flow records. Each source must be specified in an element <source>.
  If the channel receives data from all parent channels, the list of channels can be replaced with only one source: <source>*</source>. Channels in the "live" profile always have to use this notation.

Each channel within any "shadow" profile must receive from all channels from a parent profile i.e. it must always use only '*' source! It is due to the fact that for later evaluation of queries over shadow profiles (for example by fdistdump or other tools) the information about parent channels that belong to every flow has been already lost.

Optionally, each channel may contain one:

• <filter>
  A flow filter expression that will be applied to flow records received from sources. The records that satisfy the specified filter expression will be labeled with this channel and the profile to which this channel belongs. If the filter is not defined, the expression is always evaluated as true.

Warning: User must always make sure that intersection of records that belong to multiple channels of the same profile is always empty! Otherwise, the record can be stored multiple times (in case of lnfstore) or added to the summary statistic of the profile multiple times (in case of profilestats).

Example configuration

Following configuration is based on the hierarchy mentioned earlier but few parts have been simplified.

    <profile name="live">
        <type>normal</type>
        <directory>/some/directory/live/</directory>
        <channelList>
            <channel name="ch1">
                <sourceList><source>*</source></sourceList>
                <filter>odid 10</filter>
            </channel>
            <channel name="ch2">
                <sourceList><source>*</source></sourceList>
                <filter>odid 20</filter>
            </channel>
        </channelList>

        <subprofileList>
            <profile name="emails">
                <type>normal</type>
                <directory>/some/directory/emails/</directory>

                <channelList>
                    <channel name="pop3">
                        <sourceList>
                            <source>ch1</source>
                            <source>ch2</source>
                            <!-- or just <source>*</source> -->
                        </sourceList>
                        <filter>port in [110, 995]</filter>
                    </channel>

                    <!--
                    <channel name="smtp">
                        ...
                    </channel>
                    -->
                </channelList>
            </profile>

            <!--
            <profile name="buildings">
                ...
                <subprofileList>
                    <profile name="office">
                        ...
                    </profile>
                    <profile name="factory">
                        ...
                    </profile>
                </subprofileList>
            </profile>
            -->
        </subprofileList>
    </profile>

Tips

If you need to distinguish individual flow exporters, we highly recommend configuring each exporter to use unique Observation Domain ID (ODID) (IPFIX only), configure each channel of the "live" profile to represent one exporter and use filter keyword "odid" (see the filter syntax for more details and limitations). If the ODID method is not applicable in your situation, you can also use "exporterip", "exporterport", etc. keywords, but be aware, this doesn't make sense in case of the distributed collector architecture because all flows are sent to the one active proxy collector and then redistributed by the proxy to subcollectors that perform profiling. From the point of view of any subcollector the proxy is the exporter, therefore these ODID replacements don't work as expected. On the other hand, the ODID always works.

If you want to make sure that your configuration file is ready to use, you can use a tool called "ipfixcol-profiles-check". Use "-h" to show all available parameters.

Filter syntax

The filter syntax is based on the well-known nfdump tool. Although keywords must be written with lowercase letters. Any filter consists of one or more expressions expr. Any number of expr can be linked together: expr and expr, expr or expr, not expr and (expr).

An expression primitive usually consists of a keyword (a name of an Information Element), optional comparator, and a value. By default, if the comparator is omitted, equality operator = will be used. Numeric values can use scaling of following supported scaling factor: k, m, g. The factor is 1000.

Following comparators comp are supported:

• equals sign (=, == or eq)
• less than (< or lt)
• more than (> or gt)
• like/binary and (&);

Below is the list of the most frequently used filter primitives that are universally supported. If you cannot find the primitive you are looking for, try to use the corresponding nfdump expression or just use the name of IPFIX Information Element. If you need to preserve compatibility with fdistdump, you have to use only nfdump expressions!

• IP version
  ipv4 or inet4 for IPv4
  ipv6 or inet6 for IPv6

• Protocol
  proto <protocol>
proto <number>
  where <protocol> is known protocol such as tcp, udp, icmp, icmp6, etc. or a valid protocol number: 6, 17 etc.

• IP address
  [src|dst] ip <ipaddr>
[src|dst] host <ipaddr>
  with <ipaddr> as any valid IPv4 or IPv6 address. To check if an IP address is in a known IP list, use:
  [src|dst] ip in [ <iplist> ]
[src|dst] host in [ <iplist> ]
  where <iplist> is a space or comma separated list of individual <ipaddr>.

  IP addresses, networks, ports, AS number etc. can be specifically selected by using a direction qualifier, such as src or dst.

• Network
  [src|dst] net a.b.c.d m.n.r.s
  Select the IPv4 network a.b.c.d with netmask m.n.r.s.
  
  [src|dst] net <net>/<num>
  with <net> as a valid IPv4 or IPv6 network and <num> as mask bits. The number of mask bits must match the appropriate address family in IPv4 or IPv6. Networks may be abbreviated such as 172.16/16 if they are unambiguous.

• Port
  [src|dst] port [comp] <num>
  with as any valid port number. If comp is omitted, '=' is assumed.
  [src|dst] port in [ <portlist> ]
  A port can be compared against a know list, where <portlist> is a space or comma separated list of individual port numbers.

• Flags
  flags <tcpflags>
  with <tcpflags> as a combination of:
  A - ACK
  S - SYN
  F - FIN
  R - Reset
  P - Push
  U - Urgent
  X - All flags on
  The ordering of the flags is not relevant. Flags not mentioned are treated as don't care. In order to get those flows with only the SYN flag set, use the syntax flags S and not flags AFRPU.

• Packets
  packets [comp] <num> [scale]
  To filter for records with a specific packet count.
  Example: packets > 1k

• Bytes
  bytes [comp] <num> [scale]
  To filter for records with a specific byte count.
  Example: bytes 46 or bytes > 100 and bytes < 200

• Packets per second (calculated value)
  pps [comp] num [scale]
  To filter for flows with specific packets per second.

• Duration (calculate value)
  duration [comp] num
  To filter for flows with specific duration in milliseconds.

• Bits per second (calculated value)
  bps [comp] num [scale]
  To filter for flows with specific bytes per second.

• Bytes per packet (calculated value)
  bpp [comp] num [scale]
  To filter for flows with specific bytes per packet.

Following expressions are available only for processing live IPFIX records and therefore are not supported by fdistdump.

• Observation Domain ID (ODID)
  odid [comp] <number>
  To filter IPFIX records with a specific Observation Domain ID.

• Exporter IP
  exporterip <ipaddr>
exporterip in [ <iplist> ]
  To filter for exporters connected with specified IP addresses.

• Exporter port
  exporterport <port>
exporterport in [ <portlist> ]
  To filter for exporters connected with specified port.

• Collector IP
  collectorip <ipaddr>
collectorip in [ <iplist> ]
  To filter for an input IP address of a running collector.

• Collector port
  collectorport <port>
collectorport in [ <portlist> ]
  To filter for an input port of a running collector.

Instead of the identifiers above you can also use any IPFIX Information Element (IE) that is supported by IPFIXcol. These IEs can be easily added to the configuration file of the collector so even Private Enterprise IEs can be also used for filtering. See its manual page for more information. Just keep in mind that these identifiers are not supported by fdistdump right now.

For example, IPFIX IE for the source port in the transport header is called "sourceTransportPort" and essentially corresponds to the filter expression "src port". Therefore, the expressions sourceTransportPort 80 and src port 80 represent the same filter.

Filter examples

To dump all records of host 192.168.1.2:
ip 192.168.1.2

To dump all record of network 172.16.0.0/16:
net 172.16.0.0/16

To dump all port 80 IPv6 connections to any web server:
inet6 and proto tcp and ( src port > 1024 and dst port 80 )

Use-case example

Let's say that we would like to filter only POP3 flows. We know that ports of POP3 communication are 110 (unencrypted) or 995 (encrypted). So we can write:
sourceTransportPort == 110 or sourceTransportPort == 995 or destinationTransportPort == 110 or destinationTransportPort == 995

Instead of IPFIX IEs, we can use universal identifiers:
src port 110 or src port 995 or dst port 110 or dst port 995

Source and destination port can be merged:
port 110 or port 995

This expression still can be simplified:
port in [110, 995]

All examples above represent the same filter.

Back to Top