Overview of SCTP (Stream Control Transmission Protocol)

© Peter R. Egli 2016
1/45
Rev. 2.02
SCTP – Stream Control Transmission Protocol indigoo.com
INTRODUCTION TO SCTP, A GENERAL PURPOSE
TRANSPORT PROTOCOL SUITED FOR HIGH
RELIABILITY APPLICATIONS
PETER R. EGLI
INDIGOO.COM
STREAM CONTROL
TRANSMISSION PROTOCOL
SCTP

2/45
Rev. 2.02
Contents
1. What is SCTP?
2. Why use SCTP?
3. Main features of SCTP
4. The SCTP model
5. SCTP packet format
6. Association setup with SCTP
7. Association shutdown with SCTP
8. Multihoming with SCTP
9. SCTP fragmentation
10. SCTP flow control
11. SCTP congestion control
12. SCTP error control
13. Security with SCTP
14. SCTP support in different OSs and platforms
15. SCTP API in Linux (socket interface)

3/45
Rev. 2.02
1. What is SCTP?
SCTP (Stream Control Transmission Protocol, RFC4960) is a transport protocol on OSI layer 4
(like TCP or UDP).
SCTP was specifically designed as a transport protocol for public telephony network signalling
message transport. However, SCTP is generic and may supersede TCP in other applications as
well.
Simplified OSI stack (session and presentation layers omitted):
Network Layer (IP)
Data Link Layer
Physical Layer
SCTP Transport
Service
SCTP User Appl. Logical communication between
SCTP layers.
Network Layer (IP)
Data Link Layer
Physical Layer
SCTP Transport
Service
SCTP User Appl.

4/45
Rev. 2.02
2. Why use SCTP? (1/2)
SCTP has similar characteristics and applications as TCP (RFC793), but includes some
important improvements:
1. No head-of-line blocking:
TCP imposes a strict data ordering. However, if a user data message is lost during transit, all
subsequent user data messages are delayed until the lost message has been retransmitted (=
head-of-line blocking).
Some applications do not require a strict ordering of messages. E.g. applications that
exchange unrelated application messages could cope with out-of-order messages (messages
are simply processed as they arrive at the receiver). But TCP does not allow messages to pass
each other.
2. No stream-oriented data transfer:
TCP is stream-oriented. This means that TCP treats data chunks transmitted by an application
as an ordered stream of bytes (=octets in network speak). While this concept supports a wide
range of applications (message oriented like email, character oriented like TELNET, stream
oriented like video), it is unsuited in most applications because these exchange application
level messages.
SCTP preserves application level message boundaries, thus liberating applications from
implementing a framing protocol on top of the transport protocol for delineating messages.
SCTP simply maps application messages to chunks on the transmit path and back to
application messages on the receive path.

5/45
Rev. 2.02
2. Why use SCTP? (2/2)
3. Multihoming:
Since a TCP connection is defined by the quadruple source IP, destination IP, source port and
destination port, TCP does not support multihoming (use of multiple IP addresses on either
side of the connection to allow multiple transmission paths through the network thus
increasing reliability).
SCTP has built-in support for multihoming which offloads high-availability applications from
implementing this feature.
4. Certain protection against denial of service attacks:
The connection setup of TCP allows denial of service attacks, particularly SYN attacks. Each
time the TCP layer receives a SYN packet for setting up a new connection, it allocates a data
structure for storing connection parameters (called Transport Control Block). Flooding with a
high number of such SYN packets may lead to memory exhaustion.
SCTP implements a procedure to avoid or at least make it more difficult for an attacker to
launch a connection denial of service attack (4-way connection setup with state cookie).

6/45
Rev. 2.02
3. Main features of SCTP
Multiple data stream support:
Support for multiple logical streams of application messages. Ordering of messages within a
stream. Avoidance of head-of-line blocking.
Message oriented data transfer:
Transport of user data as messages, preservation of application level message boundaries.
Multihoming for network redundancy:
Use of multiple IP addresses per SCTP endpoint to allow transmission of data chunks through
different network paths.
Denial of service attack protection:
Some measures to protect against denial of service attacks such as connection setup flooding.
Fragmentation:
Detection of path MTU and fragmentation of data chunks to fit into the available path MTU.
Error correction:
Acknowledged error-free, non-duplicated data transfer.
Congestion avoidance:
Similar functionality as in TCP to avoid congestion to build up in the network.

7/45
Rev. 2.02
4. The SCTP model (1/4)
Concepts and terms of SCTP and their relation:
SCTP user
application (ULP)
U
M
Streams
SCTP socket
interface
U
M
U
M
SCTP layer
H
U
M
H
U
M
H
U
M
SCTP data chunks
H
U
M
H
U
M
H
U
M
SCTP H.
H
U
M
User
message
SCTP packet
SCTP
endpoint
IP layerTransport
address
Bundling
L1 & L2
SCTP user
application (ULP)
U
M
U
M
U
M
H
U
M
Bundling
L1 & L2
SCTP association between endpoints (logical connection)
Multiplexing /
demultiplexing of chunks
into an SCTP packet
H
U
M
H
U
M
H
U
M
SCTP H.
H
U
M
H
U
M H
U
M
UM: User Message
H: Header
L1,L2: Layer 1, Layer 2
ULP: Upper Layer Protocol

8/45
Rev. 2.02
Association:
An SCTP association is the logical relationship between 2 SCTP endpoints (≈ SCTP
connection).
Bundling:
Bundling packs multiple chunks (user data chunks and SCTP control chunks) into an SCTP
packet.
Chunk:
Unit of information within an SCTP packet. Chunks contain either user data (user data chunk)
or SCTP control information (control chunk). Each chunk has its own header (chunk header).
Endpoint:
An SCTP endpoint is an addressable logical endpoint of an SCTP association. An SCTP
endpoint represents exactly 1 SCTP port number, but may contain multiple transport addresses
in case of multihoming (1 SCTP port number, multiple IP addresses).
Stream:
A stream is a logical channel transporting in-order application messages.
Streams are unidirectional. If an application requires a bidirectional stream, it must open 1
outgoing and 1 incoming unidirectional stream and treat them together as a bidirectional
stream.

9/45
Rev. 2.02
SCTP packet:
SCTP transport PDU that contains multiple chunks and an SCTP header. The SCTP packet is
encapsulated into an IP packet.
Socket interface:
Platform specific API (OS) for opening an SCTP association and sending and receiving user
messages.
Transport address:
An SCTP transport address contains the SCTP port number and one specific IP address.
Transport control block (TCB):
Data structure containing connection information on either side of an SCTP connection,
maintained by the SCTP layer.
User application:
Application that opens an SCTP association to a peer SCTP user application and exchanges
messages with the peer application through the SCTP socket interface.
Also called ULP (Upper Layer Protocol).
User message:
Unit of data sent by the SCTP user application over the SCTP socket interface. In protocol
speek, a user message is an 'APDU' (Aplication Protocol Data Unit).

10/45
Rev. 2.02
Entity relationships (UML notation):
SCTP
socket
SCTP
Endpoint
SCTP
Association
1
2
Transport
address
IP address
1
1..*
SCTP port
number
<<represents>> 1
1
An SCTP endpoint may contain
multiple transport addresses that
each contain one IP address and
share the same SCTP port number.
Thus an SCTP endpoint has exactly
one port number.
An SCTP association is represented and defined by
the corresponding transport addresses on either
side of the association. Example:
{ [10.1.1.1 : 80], [20.1.1.1, 21.1.1.2 : 8080] }
Endpoint 1
transport
address
Endpoint 2 transport
addresses (2 IP addresses
sharing the same port number)
A specific port number
or IP address may be used
by multiple endpoints.
Only the combination of
IP address and port number
(=transport address) is
unique to an endpoint.
1
11

11/45
Rev. 2.02
5. SCTP packet format (1/3)
SCTP packet header and chunk format:
The SCTP header contains only a limited set of fields.
More SCTP control information is contained in control chunks that are bundled with user data
chunks into the SCTP packet payload.
2 Field length (bytes)2
Source port number Destination port number
Verification tag
Cecksum
Chunk #1
…
Chunk #n
SCTP common header:
Source port #: Sender's port number
Dest. port #: Receiver's port number
Verif. tag: Used by receiver to validate the sender
Checksum: CRC32 checksum over entire SCTP packet
SCTP payload:
Sequence of user data and control chunks.
Each chunk consists of a chunk header and a
chunk value (chunk format see below).
Chunk value
1 1 2
Chunk:
Chunk type: DATA or control (INIT, INITACK etc.)
Chunk flags: Chunk-specific bits
Chunk length: Size of chunk including chunk header
Chunk value: Chunk-specific data
Chunk lengthChunk type Chunk flags

12/45
Rev. 2.02
SCTP DATA chunk format:
User data chunks carry application data along with some chunk and stream management data.
User Data (sequence n of stream S)
1 1 2
Data chunk:
U bit: If set to 1, indicates that this is an unordered chunk. Unordered chunks are transmitted and sent to the
receiving application as is without re-ordering based on the sequence number.
B bit: Begin of fragment bit. If set to 1 this is the first fragment of a larger fragmented user data message.
E bit: End of fragment bit. If set to 1 this is the last fragment of a larger fragmented user data message.
Stream identifier: Identifies the stream to which this chunk belongs.
Stream seq. no.: Sequence number of user data within this stream. In case of fragmentation this numer is identical for all
fragments.
Payload proto. id.: Identifies the upper (application) layer protocol (e.g. HTTP).
User data: Application user data (e.g. HTTP header and payload).
Chunk Length (CL)Type = 0 EBU
TSN
Stream Identifier (S) Stream Sequence Number (n)
Payload Protocol Identifier

13/45
Rev. 2.02
SCTP chunk types:
SCTP packet chunks are either data (DATA) or control chunks.
These chunks may be bundled in an SCTP packet in any order.
Chunk Type ID Description
DATA 0 User payload data.
INIT 1 Used for the initiation of an SCTP association.
INIT_ACK 2 Initiation Acknowledegment.
SACK 3 Selective Acknowledegment of DATA chunks.
HEARTBEAT 4 Heartbeat request for probing reachability of a peer transport address.
HEARTBEAT_ACK 5 Heartbeat Acknowledgement.
ABORT 6
Used to immediately close an association. Subsequent control chunks are ignored by
SCTP.
SHUTDOWN 7 Used for gracefully shutting down an association.
SHUTDOWN_ACK 8 Shutdown Acknowledgement.
SHUTDOWN_COMPLETE 14 Acknowledge receipt of SHUTDOWN_ACK and completion of shutdown process.
ERROR 9
Used for reporting an error condition to the SCTP peer. An ERROR chunk may be
followed by an ABORT (fatal error).
COOKIE_ECHO 10 Chunk containing the State Cookie in the association setup.
COOKIE_ACK 11 Cookie Acknowledgement.

14/45
Rev. 2.02
6. Association setup with SCTP (1/4)
SCTP uses a 4-way handshake for establishing an association (= connection).
This scheme helps reducing vulnerability to resource attacks (akin to TCP SYN floods).
INIT chunk
* Init. tag
1
SCTP 'A' SCTP 'Z'
INIT_ACK chunk
* Cookie
2
'Z' calculates a state cookie (hashing
of a temporary TCB), but does NOT
allocate any connection resources
('Z' deletes the temporary TCB).
 Reduce vulnerability to DoS
or resource attacks.COOKIE_ECHO chunk
* Cookie
* Optionally bundled with user DATA chunks3
TCB
'Z' calculates the cookie from the
connection parameters and compares
it with the received cookie.
In case of a match, 'Z' allocates a TCB
for this new SCTP association.
COOKIE_ACK chunk
* Optionally bundled with user DATA chunks
4
TCB
Key:
TCB: Transport Control Block

15/45
Rev. 2.02
State diagram for an association setup:
CLOSED
COOKIE-
WAIT
COOKIE-
ECHOED
ESTAB-
LISHED
T: Rx INIT (passive open)
A: Create temp. TCB
Create cookie
Tx INIT_ACK
Delete temp. TCB
T: Rx ABORT (from any state)
A: Delete TCB
or
T: Abort()
A: Tx ABORT
Delete TCB
T: Associate() (active open)
A: Create TCB
Tx INIT
Start init timer
T: Rx INIT_ACK
A: Tx COOKIE_ECHO
Stop init timer
Start cookie timer
T: Rx COOKIE_ACK
A: Stop cookie timer
T: Rx valid COOKIE_ECHO
A: Create TCB
Tx COOKIE_ACK
Key:
T: Trigger causing the transition
A: Action to be carried out
Rx: Receive
Tx: Transmit
ABORT: SCTP chunks are in uppercase
Abort(): SCTP API call
ULP: Upper Layer Protocol
T: Rx INIT
A: Tx INIT_ACK
T: Rx COOKIE_ECHO
A: Tx COOKIE_ACK
T: Rx INIT (peer crashed)
A: Find existing TCB
Create temp. TCB
Tx INIT_ACK
Destroy temp. TCB
or
T: Rx COOKIE_ECHO
A: Announce restart to ULP
Reset association (counters etc.)
Tx COOKIE_ACK

16/45
Rev. 2.02
Setup collision:
Similar to TCP, an association setup collision (both SCTP 'A' and 'Z' attempt to open an
association to each other at roughly the same time) will result in a single association.
INIT chunks
* Init. tag
1
INIT_ACK chunks
* Cookie
2
COOKIE_ECHO chunks
* Cookie
3
TCBTCB
COOKIE_ACK chunks4
CLOSED
COOKIE-
ECHOED
COOKIE-
WAIT
ESTAB-
LISHED
State transitions ‘A’ & 'Z':SCTP 'A' SCTP 'Z'

17/45
Rev. 2.02
Opening an association to a closed port:
Similar to TCP where a SYN to a closed port is answered with a RST packet, SCTP sends back
an ABORT chunk. This signals to the INIT-chunk sender that the addressed port is closed.
INIT chunk
* Init. tag1
ABORT chunk2
TCP SYN segment1
TCP RST segment2
The addressed TCP
port is closed on 'Z'.
The addressed SCTP
port is closed on ‘Z’.
TCP 'A'
SCTP 'A'
TCP 'Z'
SCTP 'Z'

18/45
Rev. 2.02
7. Association shutdown with SCTP (1/2)
Unlike TCP, SCTP does not support half-open associations (half-close). A shutdown sequence
always closes all streams irrespective of their direction.
SHUTDOWN chunk1
SHUTDOWN_ACK chunk2
SHUTDOWN_COMPLETE chunk3
SHUTDOWN chunks1
SHUTDOWN_ACK chunks2
SHUTDOWN_COMPLETE chunks3
Normal shutdown procedure where
either side initiates the shutdown
by sending a SHUTDOWN chunk.
Shutdown collision: Both sides of the
association send a SHUTDOWN chunk
at around the same time.
SCTP 'A'
SCTP 'A'
SCTP 'Z'
SCTP 'Z'

19/45
Rev. 2.02
7. Association shutdown with SCTP (2/2)
State diagram for an association shutdown:
ESTAB-
LISHED
SHUTDOWN
PENDING
T: Shutdown()
A: Check outstanding
DATA chunks
SHUTDOWN
SENT
T: No more outstanding DATA chunks
A: Tx SHUTDOWN
Start shutdown timer
CLOSED
T: Rx SHUTDOWN_ACK
A: Stop shutdown timer
Tx SHUTDOWN_COMPLETE
Delete TCB
SHUTDOWN
RECEIVED
T: Rx SHUTDOWN
A: Check outstanding
DATA chunks
SHUTDOWN
_ACK SENT
T: Rx SHUTDOWN
A: Tx SHUTDOWN_ACK
T: No more outstanding
DATA chunks
A: Tx SHUTDOWN_ACK
T: Rx SHUTDOWN COMPLETE
Delete TCB
or
T: Rx SHUTDOWN_ACK
Tx SHUTDOWN_COMPLETE
Delete TCB
Key:
T: Trigger causing the transition
A: Action to be carried out
Rx: Receive
Tx: Transmit
ABORT: SCTP chunks are in uppercase
Abort(): SCTP API call

20/45
Rev. 2.02
IP 1
IP 2
8. Multihoming with SCTP
Multihoming in SCTP allows using multiple transport addresses in an association. Typically,
different transport addresses (IP addresses) are bound to different networks thus providing
resiliency in case of a network failure.
One of the transmission paths is the primary path. If connectivity over the primary path fails
(timeouts of sent packets), SCTP falls back to a secondary (alternate) path for transmitting.
N.B.: SCTP does not perform load balancing or load sharing.
The ULP can request SCTP to send data to a specific destination transport address.
Multihoming scenario with a mobile device and a primary and secondary path:
Mobile
device
Server
IP A
IP B
Network 2
Network 1
Primary path
Secondary path

21/45
Rev. 2.02
9. SCTP fragmentation
SCTP includes an (optional) fragmentation on the transmit side to avoid fragmentation in the
IP layer. An SCTP receiver must support fragmentation (be able to reassemble a fragmented
packet).
An SCTP sender must fragment DATA chunks if they do not fit into an SCTP packet including
chunk headers (exceed the association path MTU = smallest MTU of all paths to all peers in a
multihomed scenario).
SCTP user
application
300
IP Layer
2500
300 H SCTP H.1000 H
1500 H SCTP H.
300 H SCTP H.1000 H
Fragmentation of
2500 byte
user message and
bundling of DATA chunks
in the SCTP layer
SCTP user
application
300
IP Layer
2500
300 H SCTP H.1000 H
1500 H SCTP H.
IP H.1500 H SCTP H. IP H.
User messages

22/45
Rev. 2.02
SCTP User
Application
‘Z’ (ULP ‘Z’)
10. SCTP flow control (1/7)
Akin to TCP, SCTPs flow control mechanism is based on a receiver window size (rwnd).
The SCTP flow control algorithm guarantees that the receive buffer never experiences overflow
(sent data always fits into the receive buffer).
The following scenarios exemplify and explain the various algorithms and mechanisms used
in the SCTP flow control.
The examples use the following conceptual model:
SCTP Layer
‘A’
SCTP User
Application
‘A’ (ULP ‘A’)
SCTP socket
interface
Network
Transmit
buffer (queue)
Receive
buffer (queue)
SCTP Layer
‘Z’
Receive
buffer (queue)
Transmit
buffer (queue)
Application
reads
Application
writes
rwnd
‘Z’ rwnd cwnd
T3-rtx TSN
‘A’ rwnd cwnd
T3-rtx TSN
Application
writes
Application
reads
SCTP socket
interface

23/45
Rev. 2.02
Receive window (rwnd):
The recieve window size rwnd indicates the receive buffer size. The advertised receive window
size a_rwnd is the value of rwnd sent by a receiver SCTP layer to its peer STCP layer to inform
it about its receive buffer size.
A sender must not send more DATA chunks than fit into rwnd. If a_rwnd reaches zero, the
sender must stop sending DATA chunks.
Exception: 1 DATA chunk may be in flight (underway) if allowed by cwnd for window probing
(see below).
The default initial value for rwnd is 1500 bytes.
An SCTP receiver increases a_rwnd by the amount of data delivered to the ULP and advertises
new values to the SCTP transmitter.
TSN – DATA chunk counter:
Every DATA chunk has a Transmission Sequence Number (TSN). As opposed to TCP, SCTP
counts full DATA chunks and not bytes (SCTP is message oriented, not stream oriented). TSNs
are used by the receiver to acknowledge successful receipt of DATA chunks (in SACK chunks).
Congestion control window (cwnd):
The size of the congestion control window is determined by the congestion control algorithm
(see below). At any given time, the amount of outstanding data (DATA chunks in flight but not
yet acknowledged) must not exceed the congestion window size (cwnd).
If cwnd=0, the sender must stop sending SCTP packets.

24/45
Rev. 2.02
Zero window probe / window probing:
When a receiver has advertised a zero receive window (a_rwnd=0) and a subsequent SACK
advertising a non-zero receive window is lost, the sender will be blocked forever from sending
since only a new DATA chunk would trigger a new SACK packet with the non-zero rwnd
advertisement.
To overcome this potential deadlock situation, a sender may send a zero window probe after
RTO time elapsed if it received a_rwnd=0. After that, the sender should exponentially increase
the zero window probe intervals. Again, if cwnd=0, the sender must not send anything (not
even zero window probes).
A zero window probe may only be sent when all DATA chunks have been acknowledged and
no DATA chunks are inflight.
Bundling policy:
Bundling is the process of packing different chunks (DATA chunks, control chunks such as
SACK) into an SCTP packet. An SCTP sender must heed the following policies:
a. The SCTP packet size must not exceed the association path MTU, i.e. the smallest MTU of all
paths of an SCTP association (in multihomed scenarios there are multiple paths between the
SCTP peers).
b. Bundle SACK chunks with highest priority.
c. After bundling SACK chunks, if both rwnd and cwnd permit, bundle DATA chunks eligible for
retransmission into the remaining space of an SCTP packet.
d. Bundle the remaining space with new DATA chunks (again if rwnd and cwnd permit).

25/45
Rev. 2.02
Reassembly in the receiver:
An SCTP receiver first reassembles DATA chunks into the original ULP messages, then delivers
these messages to the ULP.
Silly window syndrome prevention:
The Silly Window Syndrome (SWS) can occur if both of the following are true:
a. Receive buffer almost full, receiver advertises small rwnd values.
b. Sender sends tinygrams, i.e. very small packets as allowed by rwnd.
This results in very poor transmission performance. To avoid SWS, an SCTP sender has to
apply the mechanisms set forth in RFC1122 (basically the receiver has to restrain from
advertising small rwnd values and the sender has to avoid sending tinygrams).
Transmit buffer size:
The transmit buffer in the SCTP sender may accept more data than is advertised by the SCTP
reciever with a_rwnd. An SCTP sender reports transmit buffer overflow back to the ULP by
some appropriate means (socket callback etc.). The handling of such situations is up to the
ULP (outside of scope of SCTP).
Delayed ack:
The delayed ack guidelines set forth in RFC2581 should be followed by an SCTP sender
(send acknowledge for every other SCTP DATA packet, no later than 200ms after reception of a
DATA packet).

26/45
Rev. 2.02
Flow control scenarios:
INIT [Initial-TSN=X,
a_rwnd=5000]
1
ULP ‘A’
socket()
SCTP Layer ‘Z’ ULP ‘Z’
socket()
INIT_ACK [Initial-TSN=Y,
a_rwnd=2500]
COOKIE_ECHO
COOKIE_ACK
Association
Setup
write(D=500)
rwnd 2500
T3-rtx 0
cwnd 4380
TSN X+0
rwnd 5000
T3-rtx 0
cwnd 4380
TSN Y+0
rwnd 2000
T3-rtx 0
cwnd 4380
TSN X+0
rwnd 5000
T3-rtx 0
cwnd 4380
TSN Y+0
X+0
DATA [TSN=X+0, CL=500]
2
write(D=1000) rwnd 1000
T3-rtx 0
cwnd 4380
TSN X+1
rwnd 5000
T3-rtx 0
cwnd 4380
TSN Y+0
3
X+0
X+0X+1 X+0X+1
SCTP Layer ‘A’

27/45
Rev. 2.02
Flow control scenarios:
ULP ‘A’ SCTP Layer ‘Z’ ULP ‘Z’
rwnd 1000
T3-rtx 0
cwnd 4380
TSN X+1
SACK [Cumulative TSN=X+1,
a_rwnd=1000]
4
rwnd 5000
T3-rtx 0
cwnd 4380
TSN Y+0
X+0X+1
read(D=500)
rwnd 1500
T3-rtx 0
cwnd 4380
TSN X+1
rwnd 5000
T3-rtx 0
cwnd 4380
TSN Y+0
X+1
Y+0
write(D=1000)
5
DATA [TSN=Y, CL=1000]
SACK [Cumulative
TSN=X+1, a_rwnd=1500]
Y+0
SACK [Cumulative TSN=Y,
a_rwnd=4000]
6
rwnd 500
T3-rtx 0
cwnd 4380
TSN X+4
rwnd 5000
T3-rtx 0
cwnd 4380
TSN Y+0
X+1
write(D=1000) X+2X+3
write(D=1000)
X+2
Y+0write(D=800)
X+4
SCTP Layer ‘A’

28/45
Rev. 2.02
1. SCTP association setup:
Both SCTP peers communicate their respective receive buffer size as a_rwnd (advertised receive window) in the assocation
setup. Both SCTP peers initialize their rwnd variable with the a_rwnd advertised by the peer.
2. & 3. Application writes:
ULP ‘A’ writes an application message to the SCTP layer through the socket API.
SCTP ‘A’ stores the message in the transmit buffer along with the TSN of this chunk (X) and decrements rwnd by the size of
the data (500 bytes).
SCTP ‘Z’ receives the DATA chunk and stores it in its receive buffer.
Since SCTP mandates the implementation of delayed acks as per RFC1122, SCTP ‘Z’ does not immediately send back a SACK
chunk. Instead it waits up to 200ms in order to receive another DATA chunk so as to acknowledge both chunks with one
SACK.
In the meantime, ULP ‘A’ writes another message over the API into the SCTP layer which delivers it as another DATA chunk
with TSN=X+1.
4. SACK chunk:
SCTP ‘Z’ received 2 DATA chunks before the delayed ack timer expired (200ms), so it returns a SACK chunk with cumulative
TSN = X+1 singaling that it successfully received all DATA chunks up until and including TSN=X+1.
The a_rwnd value is still 1000 because both DATA chunks X and X+1 still occupy space in the receive buffer because they
have not yet been delivered to the ULP ‘Z’.
5. ULP ‘Z’ reads and writes data:
ULP ‘Z’ reads 500 bytes from the SCTP layer and thus frees up 500 bytes in the receive buffer. Right afterwards, ULP ‘Z’ writes
a 1000 byte message to the SCTP layer which bundles it as a DATA chunk with a SACK chunk advertising a new a_rwnd value
(1500 bytes free in receive buffer) and sends it to the SCTP ‘A’ layer.
6. ULP ‘A’ writes multiple messages:
ULP ‘A’ writes two 1000 byte and one 800 byte messages to its SCTP layer. Since a_rwnd advertised by SCTP ‘Z’ is 1500, SCTP
‘A’ only sends the first 1000 byte message as a DATA chunk. In order to implement the delayed ack algorithm (RFC1122), SCTP
‘A’ bundles a SACK chunk advertising a_rwnd=4000.

29/45
Rev. 2.02
11. SCTP congestion control (1/6)
Congestion control tries avoiding overload situations in network components like routers.
Congestion in network components can lead to packet loss which is handled by the error
control function of SCTP (see below). The goal of congestion control is to avoid packet loss in
the first place.
SCTP congestion control key facts:
• SCTP congestion control is based on RFC2581 with some minor modifications.
• Congestion control is applied to the entire association, not individual streams.
• SCTP maintains a separate 𝒄𝒘𝒏𝒅 parameter for each peer destination address in
multihomed scenarios.
• As defined in RFC2581, the transmission rate starts slowly at the beginning (slow start
phase), based on feedback provided by received SACK chunks. After the slow start phase,
SCTP enters the congestion avoidance phase. In case of congestion in the network, SCTP
immediately reverts back to the slow start phase.
Slow Start
𝒄𝒘𝒏𝒅 > 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉
Congestion
Avoidance
T3-rtx timeout or fast retransmit

30/45
Rev. 2.02
SCTP congestion control window 𝐜𝒘𝒏𝒅:
𝐜𝒘𝒏𝒅 and 𝒓𝒘𝒏𝒅 (or 𝒂_𝒓𝒘𝒏𝒅 = advertised receiver window size) define 2 windows where the
smaller of the 2 determines the maximum amount of data that can be sent.
After the slow start phase, 𝐜𝒘𝒏𝒅 is large so that 𝒓𝒘𝒏𝒅 becomes the dominant window size.
𝒄𝒘𝒏𝒅
(max. amount of outstanding data,
i.e. sent but not cumulative
TSN acknowleded)
Bytes waiting to be sent
in send buffer
Bytes sent
and acknowledged
(cum. TSN ack.)
Can be
sent
anytime
TSN
𝒓𝒘𝒏𝒅
(receive window size)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
𝒇𝒍𝒊𝒈𝒉𝒕𝒔𝒊𝒛𝒆
(data «in flight»,
i.e. sent but not
acknowledged)
Congestion control window
(𝒄𝒘𝒏𝒅)
Receive window
(𝒓𝒘𝒏𝒅)
Congestion control window
in congestion avoidance phase

31/45
Rev. 2.02
Slow start and congestion avoidance phases:
𝒄𝒘𝒏𝒅𝒊𝒏𝒊𝒕 = 𝒎𝒊𝒏 𝟒 ∗ 𝑷𝑴𝑻𝑼, 𝒎𝒂𝒙 𝟐 ∗ 𝑷𝑴𝑻𝑼, 𝟒𝟑𝟖𝟎𝒃𝒚𝒕𝒆𝒔
𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉𝒊𝒏𝒊𝒕 = ∞ (𝒐𝒓 𝒂_𝒓𝒘𝒏𝒅)
∆ 𝒄𝒘𝒏𝒅 𝒔𝒔
= 𝒎𝒊𝒏 𝑫𝑨𝑻𝑨 𝒂𝒄𝒌𝒆𝒅, 𝑷𝑴𝑻𝑼
∆ 𝒄𝒘𝒏𝒅 𝒄𝒂
= 𝟏 ∗ 𝑷𝑴𝑻𝑼 𝒑𝒆𝒓 𝑹𝑻𝑻
𝑐𝑤𝑛𝑑𝑖𝑛𝑖𝑡
∆ 𝑐𝑤𝑛𝑑 𝑠𝑠
𝑠𝑠𝑡ℎ𝑟𝑒𝑠ℎ𝑖𝑛𝑖𝑡
Slow start phase
∆ 𝑐𝑤𝑛𝑑 𝑐𝑎
𝑐𝑤𝑛𝑑 𝑇3−𝑟𝑡𝑥
Congestion
avoidance phase
Slow start phase
𝒄𝒘𝒏𝒅 𝑻𝟑−𝒓𝒕𝒙 = 𝟏 ∗ 𝑷𝑴𝑻𝑼
𝒄𝒘𝒏𝒅 𝑭𝑹 = 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉 𝑭𝑹
𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉 𝑻𝟑 = 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉 𝑭𝑹 = 𝒎𝒂𝒙 ൗ𝒄𝒘𝒏𝒅
𝟐 , 𝟒 ∗ 𝑷𝑴𝑻𝑼
𝒃𝒖𝒓𝒔𝒕 𝒎𝒂𝒙 = 4
𝑠𝑠𝑡ℎ𝑟𝑒𝑠ℎ 𝐹𝑅
𝑐𝑤𝑛𝑑 𝐹𝑅
𝑠𝑠𝑡ℎ𝑟𝑒𝑠ℎ 𝑇3
Packet loss
(T3-rtx timeout) Packet loss
(Fast Retransmit)
Congestion
avoidance phase
1 2
3
SS
phase
CA
phase
4 5
6
Key:
PMTU: Association Path MTU
SS: Slow Start
CA: Congestion Avoidance
ssthresh
cwnd

32/45
Rev. 2.02
General mechanism of congestion control:
The general mechanism applied in SCTP congestion control (as per RFC2581) is to slowly increase the congestion window
size 𝒄𝒘𝒏𝒅, but to rapidly collapse the window when there are signs of congestion.
Packet loss is deemed a sign of congestion. Note, however, that this is not always true (e.g. on wireless links there may be
packet loss due to radio signal interferences).
Congestion control is applied to the entire association, not individual streams. Nevertheless, 𝒄𝒘𝒏𝒅 is maintained per
destination transport address (multihomed scenarios).
Maximum burst limitation:
At any instance of time, the maximum burst limit is 4 (a maximum of 4 packets may be sent at any opportunity to send).
This may be accomplished by regulating 𝒄𝒘𝒏𝒅 as follows:
𝒊𝒇 𝒇𝒍𝒊𝒈𝒉𝒕𝒔𝒊𝒛𝒆 + 𝒃𝒖𝒓𝒔𝒕 𝒎𝒂𝒙 ∗ 𝑷𝑴𝑻𝑼 < 𝒄𝒘𝒏𝒅 𝒕𝒉𝒆𝒏 𝒄𝒘𝒏𝒅 = 𝒇𝒍𝒊𝒈𝒉𝒕𝒔𝒊𝒛𝒆 + 𝒃𝒖𝒓𝒔𝒕 𝒎𝒂𝒙 ∗ 𝑷𝑴𝑻𝑼
Congestion control parameters:
The following parameters are involved in congestion control:
Parameter Description
𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉
Slow start threshold (threshold between slow start and congestion avoidance phase).
If 𝒄𝒘𝒏𝒅 ≤ 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉, SCTP is in the slow start phase. If 𝒄𝒘𝒏𝒅 > 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉, SCTP is in the
congestion avoidance phase.
𝒄𝒘𝒏𝒅
Congestion window size. Maximum permissible number of outstanding bytes (sent data which
are not yet acknowledged).
𝒇𝒍𝒊𝒈𝒉𝒕𝒔𝒊𝒛𝒆 Actually sent data which are not yet acknowledged.
𝒑𝒃𝒂
Partially bytes acknowledged.
Total number of bytes acknowledged including SACK gap ack blocks.

33/45
Rev. 2.02
1. Slow Start Phase:
Before any data transfer takes place, 𝒄𝒘𝒏𝒅 is initialized to 𝑐𝑤𝑛𝑑𝑖𝑛𝑖𝑡.
𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉 is initialized to some arbitrarily large value, e.g. the receiver window size (𝒂_𝒓𝒘𝒏𝒅).
The slow start phase is defined by 𝒄𝒘𝒏𝒅 ≤ 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉.
𝒄𝒘𝒏𝒅 is incremented by ∆ 𝑐𝑤𝑛𝑑 𝑠𝑠
when:
a. Current 𝒄𝒘𝒏𝒅 is fully utilized
b. Incoming SACK advances the cumulative TSN ack point
c. Sender is not in fast recovery
The slow start phase is a self clocking mechanism in that only successfully received SACK
chunks increase 𝒄𝒘𝒏𝒅. In case of congestion (packet loss), SACKs will not be received thus
𝒄𝒘𝒏𝒅 is not increased.
2. Congestion Avoidance Phase:
The congestion avoidance phase is defined by 𝒄𝒘𝒏𝒅 > 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉.
In this phase, 𝒄𝒘𝒏𝒅 is incremented linearly by 𝑷𝑴𝑻𝑼 as follows:
𝒑𝒃𝒂 = 𝟎 (initial 𝒑𝒃𝒂).
Increment 𝒑𝒃𝒂 by the total amount of data bytes acknowledged in SACK chunks.
If 𝒑𝒃𝒂 ≥ 𝐜𝐰𝒏𝒅:
𝐜𝐰𝒏𝒅 = 𝐜𝐰𝒏𝒅 + 𝑷𝑴𝑻𝑼
𝒑𝒃𝒂 = 𝒑𝒃𝒂 − 𝐜𝐰𝒏𝒅
When all sent data sent has been acknowledged by receiver: 𝒑𝒃𝒂 = 𝟎.

34/45
Rev. 2.02
3. & 4. Packet loss (retransmission timeout) with ensuing slow start phase:
If the retransmission timer T3-rtx times out due to packet loss,
𝒄𝒘𝒏𝒅 is immediately reduced to 𝑐𝑤𝑛𝑑 𝑇3−𝑟𝑡𝑥. This limits the number of sent packets in bursts
which in turn alleviates congestion in the network.
This way, SCTP reverts back to the slow start phase.
𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉 is reduced to half the last known good 𝒄𝒘𝒏𝒅 or 𝟒 ∗ 𝐏𝐌𝐓𝐔, whichever is larger. This
makes SCTP enter the subsequent congestion avoidance phase earlier, i.e. SCTP stops
incrementing 𝒄𝒘𝒏𝒅 exponentially earlier, thus again alleviating the congestion.
5. Congestion Avoidance Phase:
Again, once 𝒄𝒘𝒏𝒅 > 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉, SCTP enters the congestion avoidance phase.
6. Packet loss with fast retransmit:
Fast retransmit indicates slight congestion in the network.
Again, 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉 is reduced to half the last known good 𝒄𝒘𝒏𝒅 or 𝟒 ∗ 𝐏𝐌𝐓𝐔, whichever is larger.
In such situations, the congestion control algorithm reacts more gently in that 𝒄𝒘𝒏𝒅 is cut in
half and not down to 𝐏𝐌𝐓𝐔.
Subsequently, SCTP enters the slow start phase for a short period of time (slow start phase if
𝒄𝒘𝒏𝒅 ≤ 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉).
As soon as 𝒄𝒘𝒏𝒅 > 𝒔𝒔𝒕𝒉𝒓𝒆𝒔𝒉, SCTP enters the congestion avoidance phase again.

35/45
Rev. 2.02
12. SCTP error control (1/8)
Retransmissions of lost packets:
Packets may be lost due to buffer overflows in transit or due to bit errors in the packet itself.
SCTP handles lost packets through retransmission (akin to TCP).
A retransmission timer is started for each chunk (see procedure below).
SCTP maintains a separate retransmission timer RTO for each peer transport address.
The calculation of RTO closely follows the definitions of the TCP retransmission timer value.
If the aggregate bandwidth temporarily exceeds
the available bandwidth of an output link, the
router drops excess packets.
Packets dropped
by router 3.
Router 1
Router 2
Router 3
SCTP silently discards packets with
an incorrect checksum (CRC32).
SCTP

36/45
Rev. 2.02
Retransmission procedure:
Restart retransmission
timer (T3-rtx) with
current RTO
Yes
Retransmission
timer for peer
transport address
running?
No
Receive SACK chunk
Last outstanding
DATA chunk
acknowledged?
Stop retransmission
timer (T3-rtx)
Yes
New DATA
chunk to be
sent
No
RTORTO*2
(back off the timer)
Retransm.
timer expiry
Retransmit as many
outstanding DATA
chunks in a single SCTP
packet as permitted by
MTU
Restart retransmission
timer (T3-rtx) with
current RTO
Receive SACK chunk
Recalculate RTO

37/45
Rev. 2.02
Retransmission timer value:
SCTP maintains a retransmission timer RTO for each peer transport address.
Since the actual round trip time is subject to constant changes, it needs to be constantly
adjusted.
𝑺𝑹𝑻𝑻 𝒏𝒆𝒘 = 𝟏 − 𝑹𝑻𝑶 𝜶 ∗ 𝑺𝑹𝑻𝑻 𝒐𝒍𝒅 + 𝑹𝑻𝑶 𝜶 ∗ 𝑹′
𝑹𝑻𝑻𝑽𝑨𝑹 𝒏𝒆𝒘 = 𝟏 − 𝑹𝑻𝑶 𝜷 ∗ 𝑹𝑻𝑻𝑽𝑨𝑹 𝒐𝒍𝒅 + 𝑹𝑻𝑶 𝜷 ∗ 𝑺𝑹𝑻𝑻 𝒐𝒍𝒅 − 𝑹′
𝑹𝑻𝑶 𝒏𝒆𝒘 = 𝑺𝑹𝑻𝑻 𝒏𝒆𝒘 + 𝟒 ∗ 𝑹𝑻𝑻𝑽𝑨𝑹 𝒏𝒆𝒘
where
𝑹𝑻𝑻𝑽𝑨𝑹 = 𝒄𝒖𝒓𝒓𝒆𝒏𝒕 𝑹𝒐𝒖𝒏𝒅 𝑻𝒓𝒊𝒎 𝑻𝒊𝒎𝒆 𝒗𝒂𝒓𝒊𝒂𝒕𝒊𝒐𝒏
𝑺𝑹𝑻𝑻 = 𝒄𝒖𝒓𝒓𝒆𝒏𝒕 𝑺𝒎𝒐𝒐𝒕𝒉𝒆𝒅 𝑹𝒐𝒖𝒏𝒅 𝑻𝒓𝒊𝒑 𝑻𝒊𝒎𝒆
𝑹𝑻𝑶 = 𝒄𝒖𝒓𝒓𝒆𝒏𝒕 𝑹𝒆𝒕𝒓𝒂𝒏𝒔𝒎𝒊𝒔𝒔𝒊𝒐𝒏 𝑻𝒊𝒎𝒆𝑶𝒖𝒕 𝒕𝒊𝒎𝒆𝒓 𝒗𝒂𝒍𝒖𝒆
𝑹𝑻𝑶 𝜶 = 𝒘𝒆𝒊𝒈𝒉𝒊𝒏𝒈 𝒇𝒂𝒄𝒕𝒐𝒓 (𝒅𝒆𝒇𝒂𝒖𝒍𝒕 𝒗𝒂𝒍𝒖𝒆 = 𝟏/𝟖)
𝑹𝑻𝑶 𝜷 = 𝒘𝒆𝒊𝒈𝒉𝒊𝒏𝒈 𝒇𝒂𝒄𝒕𝒐𝒓 (𝒅𝒆𝒇𝒂𝒖𝒍𝒕 𝒗𝒂𝒍𝒖𝒆 = 𝟏/𝟒)
𝑹′
= 𝒄𝒖𝒓𝒓𝒆𝒏𝒕 𝒎𝒆𝒂𝒔𝒖𝒓𝒆𝒅 𝒓𝒐𝒖𝒏𝒅 𝒕𝒓𝒊𝒑 𝒕𝒊𝒎𝒆
RTO1 is too small. Many packets experience
round trip times larger than RTO1 which would result
in inadvertent retransmissions.
RTO2 is a good choice since it is slightly larger
than the largest round trip time but still as small
as possible.
RTO is recalculated factoring in the
currently smoothed («averaged»)
round trip time and its variation.
The weighing factors (𝑹𝑻𝑶 𝜶 and 𝑹𝑻𝑶 𝜷)
determine the speed of adjustment
of 𝑹𝑻𝑶 to new measurement
values for the round trip time (𝑹′
).
10 20 30 40 50 60
Probability
Round trip time
[milliseconds]
0.1
0.2
0.3
0.4
RTO1 RTO2
Mean value
(SRTT)
RTTVAR

38/45
Rev. 2.02
Fast Retransmit:
SCTP employs a similar mechanism for fast retransmits as TCP as shown below.
DATA [TSN=10]1
Received DATA chunks (TSN)
DATA [TSN=11,12]2
SACK [Cum. TSN Ack=12]3
DATA [TSN=13]4
DATA [TSN=14,15]5
SACK [Cum. TSN Ack=12,
Gap Ack Block # 1 Start / End = 2 / 3]
6
DATA [TSN=16]7
DATA [TSN=17]8
Miss indication for each TSN
10 - - - - - - - -
10 11 12 - - - - - -
10 11 12 - 14 15 - - -
10 11 12 - 14 15 - 17 -
0 - - - - - - - -
10 - - - - - - - -
Lost!
Lost!
0 0 0 - - - - - -
10 11 12 - - - - - -
0 0 0 - - - - - -
10 11 12 - - - - - -
0 0 0 0 - - - - -
10 11 12 13 - - - - -
0 0 0 0 0 0 - - -
10 11 12 13 14 15 - - -
0 0 0 1 0 0 - - -
10 11 12 13 14 15 - - -
0 0 0 1 0 0 - - -
10 11 12 13 14 15 - - -
0 0 0 1 0 0 0 - -
10 11 12 13 14 15 16 - -
0 0 0 1 0 0 0 0 -
10 11 12 13 14 15 16 17 -
SCTP 'A' SCTP 'Z'

39/45
Rev. 2.02
Fast Retransmit:
SCTP employs a similar mechanism for fast retransmits as TCP as shown below.
SACK [Cum. TSN Ack=12
Gap Ack Block # 1 Start / End=2 / 3
Gap Ack Block # 2 Start / End=5 / 5]
9
DATA [TSN=18]10
Gap Ack Block # 1 Start / End = 2 / 3
11
Received DATA chunks (TSN)
Retransmission, DATA [TSN=13]12
13
Retransmission, DATA [TSN=16]14
SACK [Cum. TSN Ack=18]15
10 11 12 - 14 15 - 17 18
10 11 12 13 14 15 - 17 18
10 11 12 13 14 15 16 17 18
0 0 0 2 0 0 1 0 -
10 11 12 13 14 15 16 17 -
0 0 0 2 0 0 1 0 0
10 11 12 13 14 15 16 17 18
0 0 0 3 0 0 2 0 0
10 11 12 13 14 15 16 17 18
0 0 0 0 0 0 2 0 0
10 11 12 13 14 15 16 17 18
0 0 0 0 0 0 3 0 0
10 11 12 13 14 15 16 17 18
0 0 0 0 0 0 0 0 0
10 11 12 13 14 15 16 17 18
0 0 0 0 0 0 0 0 0
10 11 12 13 14 15 16 17 18
SCTP 'A' SCTP 'Z'

40/45
Rev. 2.02
Fast Retransmit:
Step by step explanation of a Fast Retransmit procedure.
1. & 2. DATA chunks:
SCTP ‘A’ sends DATA chunks with TSNs 10, 11 and 12 in 2 separate SCTP packets. Both packets are correctly received by
SCTP ‘Z’.
3. SACK chunk:
Following the guidelines set forth in RFC2581 (Delayed Ack: Acknowledge for every other SCTP DATA packet, no later than
200ms after reception of a DATA packet), SCTP ‘Z’ sends back a SACK chunk acknowledging the correct
reception of DATA chunks 10, 11, and 12 (Cumulative TSN Ack equals 12 thus acknowledging correct reception of all TSNs up
until and including DATA chunk 12).
4. Lost SCTP packet:
The SCTP packet with TSN=13 is lost. Such packet losses are typical of routers experiencing slight congestion on their output
buffers where they drop a few excess packets.
5. Successful delivery of next DATA chunk:
SCTP DATA chunks with TSNs 14 and 15 are again successfully received by SCTP ‘Z’.
6. SCTP ‘Z’ signals missing TSNs:
SCTP ‘Z’ sends back a SACK with cumulative TSN Ack = 12 (all TSNs including 12 correctly received) and a gap ack block
with start=2 and end=3 indicating that TSN=14 (12+2) and TSN=15 (12+3) have been correctly received. This in turn signals to
SCTP ‘A’ that TSN=13 has been lost (gap). SCTP ‘A’ increments the miss indication counter for TSN=13.
7. Lost SCTP packet again:
The SCTP packet with TSN=16 is lost.
8. Successful reception of next DATA chunk:
The DATA chunk with TSN=17 is correctly received again by SCTP ‘Z’.

41/45
Rev. 2.02
Fast Retransmit:
Step by step explanation of a Fast Retransmit procedure.
9. SCTP ‘Z’ signals missing TSNs:
SCTP ‘Z’ sends back another SACK, now with 2 gap ack blocks.
The first gap ack block acknowledges the reception of TSN=14 and TSN=15, the second gap ack block acknowledges the
reception of TSN=17 (12+5). SCTP ‘A’ increments the miss indication counters for TSN=13 and TSN=16.
10. Successful reception of next DATA chunk:
The DATA chunk with TSN=18 is correctly received again by SCTP ‘Z’.
11. SACK including acknowledgement for TSN=18:
SCTP ‘Z’ sends back another SACK with 2 gap ack blocks, the first with the same start and end numbers as in step 9, the
second with start=5 (12+5) and end=6 (12+6). SCTP ‘A’ again increments the miss indication counters for TSN=13 and TSN=16.
12. Retransmission of TSN=13:
The miss indication counter for TSN=13 reaches 3, thus TSN=13 is eligible for retransmission. SCTP ‘A’ retransmits the DATA
chunk with TSN=13 which is correctly received by SCTP ‘Z’.
13. & 14. SACK with Cumulative Ack TSN=15 and retransmission of TSN=16:
SCTP ‘Z’ signals the correct reception of all DATA chunks up until and including TSN=15. The gap ack block with start=2 (15+2)
and end=3 (15+3) indicates that TSN=17 and TSN=18 were successfully received and thus that TSN=16 is still missing. SCTP
‘A’ increments the miss indication counter for TSN=16 which reaches 3 thus triggering the retransmission of TSN=16.
15. Last SACK of fast retransmission procedure:
Finally, SCTP ‘Z’ sends back a SACK signalling the correct reception of all DATA chunks up until and including TSN=18.
SCTP ‘Z’ reverts to delayed acknowledge mode (RFC2581) and SCTP ‘A’ exits the fast retransmit mode.

42/45
Rev. 2.02
SCTP uses a CRC32 checksum as per RFC3309 to provide data integrity of SCTP packets.
Fill in proper verification
tag in common header
Initialize checksum field
with ‘0’ bits
Send SCTP packet
New SCTP
chunk(s) to
be sent
Calculate CRC32 of the
whole SCTP packet (incl.
common header)
SCTP sender CRC32 handling: SCTP receiver CRC32 handling:
Store CRC32 in SCTP
common header in
temporary variable
New SCTP
chunk(s)
received
Replace CRC32 field in
common header with all
‘0’ bits
Calculate CRC32 of the
whole SCTP packet (incl.
common header)
Yes
Silently discard
SCTP packet
Process
SCTP packet
No
CRCs
match?

43/45
Rev. 2.02
13. Security with SCTP
Security (encryption, authentication etc.) was not initially designed into SCTP.
2 first approaches (SCTP over IPSec and TLS over IPSec) have limitations and are only stop-
gap solutions.
Recent activities in integrating security into the SCTP protocol itself are funneled into an IETF
draft and may eventually become a standard.
Security Protocol Standard Description
SCTP over IPSec RFC3354
• Does not support all SCTP features (no multihoming).
• 1 IPSec Security Association per IP address needed (scalability issue).
TLS over SCTP RFC3436
• Does not support all SCTP features (e.g. no unordered delivery).
• Control chunks and SCTP header not secured, only DATA chunks.
• 1 TLS connection per SCTP stream (scalability issue).
• Each DATA chunk (ULP message) individually secured (performance issue).
Secure SCTP (S-SCTP) IETF Draft
• No TLS or IPSec needed.
• Security built into the SCTP protocol.
• Fully compatible with RFC4960.

44/45
Rev. 2.02
14. SCTP support in different OSs and platforms
There is varying support for SCTP by the different OSs and platforms.
Where native (kernel) support is not available, user space SCTP libraries such as sctplib
(Linux, FreeBSD, Mac OS X, Solaris, Windows) or SctpDrvcan (Windows) can be used instead.
OS / Platform
SCTP
Support
Description
Linux Yes
SCTP supported since kernel 2.4.
lksctp-tools is a user space library that provides a specific SCTP socket
interface. Without lksctp-tools, only the standard (SCTP agnostic) socket
interface is available.
FreeBSD / OpenBSD Yes Since version 7.0.
Mac OSX Yes Since OS X version 10.7 Lion.
Solaris Yes Since Solaris 10.
Windows No
No Windows version including Windows 7, 8 and Windows RT supports SCTP.
Microsoft does not have any plans to add native SCTP (lack of customer demand
they say).
.Net No No support for SCTP in .Net framework including version 4.5.
Java (Yes)
API available under com.sun.nio.sctp since Java 7.
Requires native SCTP (kernel) support by the underlying OS.
Cisco IOS Yes Since IOS 12.
QNX Neutrino Real Time OS Yes Since 6.3.0.
Android Yes Needs the activation of the lksctp libary in the Linux kernel.

45/45
Rev. 2.02
15. SCTP API in Linux (socket interface)
On Linux, 2 APIs are available for using SCTP sockets.
1. Standard socket interface
• Linux header file: include/sys/socket.h
• listenSock = socket(AF_INET, SOCK_STREAM, IPPROTO_SCTP);
2. Socket extensions for SCTP:
• Defined in RFC6458
• Linux header file: include/net/sctp/sctp.h
SCTP
Standard Socket API
(include/sys/socket.h)
SCTP user
application (ULP)
L3
L1 & L2
SCTP Socket API
(include/net/sctp/sctp.h)
SCTP specific API.
Full support of all SCTP features.
Standard socket features only.
SCTP features like streams not available.

Overview of SCTP (Stream Control Transmission Protocol)

More Related Content

What's hot (20)

Similar to Overview of SCTP (Stream Control Transmission Protocol) (20)

More from Peter R. Egli (20)

Recently uploaded (20)

Overview of SCTP (Stream Control Transmission Protocol)