Introduction to Stream Control Transmission Protocol
Most people who have written networking software are familiar with the TCP and UDP protocols. These are used to connect distributed applications and allow messages to flow between them. These protocols have been used successfully to build Internet applications as we know them: e-mail, HTTP, name services and so forth. But, these protocols are more than 20 years old, and over time, some of their deficiencies have become well known. Although there have been many attempts to devise new general-purpose transport protocols above the IP layer, only one so far has received the blessing of the IETF: SCTP (Stream Control Transmission Protocol). The central motivation behind SCTP is to provide a more reliable and robust protocol than either TCP or UDP that can take advantage of features such as multihoming.
SCTP is not a radical departure from TCP or UDP. It borrows from both but is most similar to TCP. It is a reliable session-oriented protocol, like TCP. It adds new features and options and allows finer control over the transport of packets. In all but the “edge” cases, it can be used as a drop-in in place of TCP. This means that TCP applications often can be ported trivially to SCTP. Of course, to benefit properly from the new features of SCTP, you need to use the additional API calls for SCTP.
The first additional feature in SCTP is better support for multihomed devices—that is, computers with more than one network interface. At one time this meant only routers and bridges connecting different parts of the Internet, but now even computers on the edges of the network can be multihomed. Most laptops have built-in Ethernet cards and Wi-Fi cards, and many have Bluetooth cards as well (which have IP support through the Bluetooth PPP stack). Some laptops now are shipping with WiMAX cards, and it even is possible to run IP over the infrared port! So, the standard laptop is at least dual-homed, with possibly up to five distinct IP network interfaces.
TCP and UDP allow use of only one or all of the interfaces. But, what if you are running your laptop as a peer in, say, a file-sharing service? It probably would be silly to use the Bluetooth and infrared interfaces. WiMAX can be very expensive to shift large amounts of data. But, it would make sense to use both the Ethernet and Wi-Fi interfaces. SCTP can support this selective choosing of interfaces. Some implementations even can add and drop interfaces dynamically, so as you unplug your laptop and move out of the house, an application can switch to the WiMAX interface if you want.
The second main new feature is multistreaming—that is, one “association” (which is renamed from “connection” from TCP) can support multiple data streams. It is no longer necessary to open up multiple sockets; instead, a single socket can be used for multiple streams to a connected host. Several TCP applications could benefit from this. For example, FTP (the major file transfer protocol) uses two streams: one on port 21 for control messages and another on port 20 for data. This caused problems with firewalls in place. A client could connect to a server through a firewall, but the server could not connect to the client for data transfer because of the firewall. The FTP protocol had to be extended to allow for “passive” connections to overcome this. There would be no need for such an extension under SCTP—simply send the data on a separate stream in an association established by a client.
The X Window System also uses multiple sockets on multiple ports. Although it is not common, a computer can have multiple display devices. Typically, the first is on port 6000, the second on port 6001 and so on. Under SCTP, these could all be separate streams on a single association. HTML documents often contain embedded references to image files, and to display a page properly requires downloading the original page and all of these images (or embedded frames too). HTTP originally used a separate TCP connection per downloaded URL, which was expensive and time consuming. HTTP 1.1 brought in “persistent connections”, so that a single socket could be reused for all of these sequential downloads. Under SCTP, the separate images could be downloaded concurrently in separate streams on a single association.
There are even more subtle uses of SCTP multiple streams. An MPEG movie consists of different types of frames: I frames, P frames and B frames. I frames encode complete images, and the other two types measure differences between frames. Typically, there is an I frame every ten frames, with the others “predicted” from these. It is critical that the I frames be delivered, but less so for the P and B frames. Although SCTP is not designed as a Quality-of-Service protocol, it does allow different delivery parameters on different streams within an association, so that the I frames can be delivered more reliably.
SCTP has many more features, such as:
TCP is a byte-oriented protocol, and UDP is message-oriented. The majority of applications are message-oriented, and applications using TCP have to jump through hoops, such as sending the message length as a first parameter. SCTP is message-oriented, so such tricks are not so necessary.
A single socket can support multiple associations—that is, a computer can use a single socket to talk to more than one computer. This is not multicast, but it could be useful in peer-to-peer situations.
SCTP has no “out of band” messages, but a large number of events can be interleaved onto a single association, so that an application can monitor the state of the association (for example, when the other end adds another interface to the association).
The range of socket options is greater than TCP or UDP. These also can be used to control individual associations or individual streams within a single association. For example, messages on one stream can be given a longer time-to-live than messages on other streams, increasing the likelihood of their delivery.
|Updates from LinuxCon and ContainerCon, Toronto, August 2016||Aug 23, 2016|
|NVMe over Fabrics Support Coming to the Linux 4.8 Kernel||Aug 22, 2016|
|What I Wish I’d Known When I Was an Embedded Linux Newbie||Aug 18, 2016|
|Pandas||Aug 17, 2016|
|Juniper Systems' Geode||Aug 16, 2016|
|Analyzing Data||Aug 15, 2016|
- Updates from LinuxCon and ContainerCon, Toronto, August 2016
- NVMe over Fabrics Support Coming to the Linux 4.8 Kernel
- What I Wish I’d Known When I Was an Embedded Linux Newbie
- New Version of GParted
- All about printf
- Analyzing Data
- Tor 0.2.8.6 Is Released
- Blender for Visual Effects
- Better Cloud Storage with ownCloud 9.1
With all the industry talk about the benefits of Linux on Power and all the performance advantages offered by its open architecture, you may be considering a move in that direction. If you are thinking about analytics, big data and cloud computing, you would be right to evaluate Power. The idea of using commodity x86 hardware and replacing it every three years is an outdated cost model. It doesn’t consider the total cost of ownership, and it doesn’t consider the advantage of real processing power, high-availability and multithreading like a demon.
This ebook takes a look at some of the practical applications of the Linux on Power platform and ways you might bring all the performance power of this open architecture to bear for your organization. There are no smoke and mirrors here—just hard, cold, empirical evidence provided by independent sources. I also consider some innovative ways Linux on Power will be used in the future.Get the Guide