1.. _congestion-handling:
2
3*******************
4Congestion Handling
5*******************
6
7.. _congestion-handling-background:
8
9What is Congestion?
10===================
11
12Congestion occurs when servers are subjected to client queries faster
13than those queries can be processed. As a result, the servers begin accumulating
14a backlog of pending queries. The longer the high rate of traffic
15continues, the farther behind the servers fall. Depending on the client
16implementations, those that fail to get leases either give up or simply
17continue to retry forever. In the former case, the server may eventually
18recover, but the latter case is a vicious cycle from which the server is
19unable to escape.
20
21In a well-planned deployment, the number and capacity of servers is
22matched to the maximum client loads expected. As long as capacity is
23matched to load, congestion does not occur. If the load is routinely too
24heavy, then the deployment needs to be re-evaluated. Congestion
25typically occurs when there is a network event that causes overly large
26numbers of clients to simultaneously need leases, such as recovery after
27a network outage.
28
29The goal of congestion handling is to help servers mitigate the peak in
30traffic by fulfilling as many of the most relevant requests as possible
31until the congestion subsides.
32
33Prior to Kea 1.5, kea-dhcp4 and kea-dhcp6 read inbound packets directly
34from the interface sockets in the main application thread, which meant
35that packets waiting to be processed were held in socket buffers
36themselves. Once these buffers filled, any new packets were discarded.
37Under swamped conditions, the servers ended up processing client packets
38that were no longer relevant, or worse, were redundant. In other words,
39the packets waiting in the FIFO socket buffers became increasingly
40stale.
41
42.. _congestion-handling-solution:
43
44Configuring Congestion Handling
45===============================
46
47Kea 1.5 introduced the Congestion Handling feature. Congestion handling
48offers the ability to configure the server to use a separate thread to
49read packets from the interface socket buffers. As the thread reads
50packets from the buffers, they are added to an internal packet queue,
51and the server's main application thread processes packets from this
52queue rather than from the socket buffers. By structuring it this way, a
53configurable layer has been introduced which can make decisions on which
54packets to process, how to store them, and the order in which they are
55processed by the server.
56
57The default packet queue implementation for both kea-dhcp4 and kea-dhcp6
58is a simple ring buffer. Once it reaches capacity, new packets get added
59to the back of the queue by discarding packets from the front of the
60queue. Rather than always discarding the newest packets, Kea now always
61discards the oldest packets. The capacity of the buffer, i.e. the maximum
62number of packets the buffer can contain, is configurable. A reasonable
63starting point would be to match the capacity to the number of leases
64per second a specific installation of Kea can handle. Please note that this
65figure varies widely depending on the specifics of an individual deployment.
66
67As there is no one algorithm that will best handle the dynamics of all
68sites, and because over time new approaches will evolve, the packet
69queue is implemented as a plug-in, which can be replaced by a custom queue
70implementation via a hook library. This should make it straightforward
71for interested parties to experiment with their own solutions.
72(Developers can refer to isc::dhcp::PacketQueue and
73isc::dhcp::PacketQueueMgr, described in the
74`Kea Developer's Guide <https://reports.kea.isc.org/dev_guide/index.html>`__.)
75
76Packet queue behavior is configured in both kea-dhcp4 and kea-dhcp6
77servers through an optional, top-level, configuration element,
78``dhcp-queue-control``. Omitting this element disables packet queueing:
79
80::
81
82      "dhcp-queue-control": {
83          "enable-queue": true|false,
84          "queue-type": "queue type",
85          "capacity" : n
86      }
87
88where:
89
90-  ``enable-queue`` true|false - enables or disables packet queueing.
91   When true, the server processes packets from the packet queue, which
92   is filled by a separate thread. When false, the server processes
93   packets directly from the socket buffers in the main thread. It is
94   disabled by default.
95
96-  ``queue-type`` - name of the queue implementation to use. This value
97   exists so that custom implementations can be registered (via a hook
98   library) and then selected. There is a default packet queue
99   implementation that is pre-registered during server start up:
100   "kea-ring4" for kea-dhcp4 and "kea-ring6" for kea-dhcp6.
101
102-  ``capacity`` = n [packets] - this is the maximum number of packets the
103   queue can hold before packets are discarded. The optimal value for
104   this is extremely site-dependent. The default value is 64 for both
105   kea-ring4 and kea-ring6.
106
107The following example enables the default packet queue for kea-dhcp4,
108with a queue capacity of 250 packets:
109
110::
111
112   "Dhcp4":
113   {
114       ...
115      "dhcp-queue-control": {
116          "enable-queue": true,
117          "queue-type": "kea-ring4",
118          "capacity" : 250
119       },
120       ...
121   }
122
123The following example enables the default packet queue for kea-dhcp6,
124with a queue capacity of 300 packets:
125
126::
127
128   "Dhcp6":
129   {
130       ...
131      "dhcp-queue-control": {
132          "enable-queue": true,
133          "queue-type": "kea-ring6",
134          "capacity" : 300
135       },
136       ...
137   }
138
139.. note:
140
141   Currently the congestion handling is incompatible with multi-threading:
142   when both are enabled the congestion handling is silently disabled.
143