Consider the example of a download stream that can be throttled (eg. torrent client, dropbox sync, etc). How does a program apply backpressure to the network?
My thoughts are that, from a software perspective you can choose to read from a socket at a certain speed. But how does the socket you're reading from know that you only want your device to receive data so quickly? Does the actual NIC apply backpressure over the network somehow? If so, by what mechanism?
Backpressure is embedded in TCP/IP protocol. If slow consumer does not read bytes from connection in timely manner, producer is unable to put more bytes than there are buffer memory on sending and receiving sides.
In contrast, UDP messages are not counted and can be dropped if there is no free memory on receiver side to store them.
I am told to increase TCP buffer size in order to process messages faster.
My Question is, no matter what buffer i am using for TCP message(ByteBuffer, DirectByteBuffer etc), whenever CPU receives interrupt from say NIC, to handle network request to read the socket data, does OS maintain any buffer in memory outside Address Space of requesting process(i.g. the process which is listening on that socket)
or
whatever way CPU receives network data, it will always be written in a buffer of process address space only and no buffer(including 'Recv-Q' and 'Send-Q' of netstat command) outside of the address space is maintained for this communication?
The process by which the Linux network stack receives data is a bit complicated. I wrote a comprehensive guide to the Linux network stack that explains everything you need to know starting from the device driver up to a userland program's socket receive queue.
There are many places buffers are maintained in the kernel:
The DMA ring where packets are written by the NIC after they've arrived.
References to the packets on the DMA ring are used to process the packet.
Eventually, the packet data is added to process' receive queue, if the receive queue is not full already.
Reads from the socket will pull packets from the process' receive queue.
If packet sniffing is occurring, packet data is duplicated and sent to any filters added by the packet sniffing code.
The full process of how data is moved, accounted for, and dropped (when required) is described in the blog post linked above.
Now, if you want to process messages faster, I assume you mean you want to reduce your packet processing latency, correct? If so, you should consider using SO_BUSYPOLL which can help reduce packet processing latency.
Increasing the receive buffer just increases the number of packets that can be queued for a userland socket. To increasing packet processing power, you need to carefully monitor and tune each component of the network stack. You may need to use something like RPS to increase the number of CPUs processing packets.
You will also want to monitor each component of your network stack to ensure that available buffers and CPU processing power is sufficient to handle your packet workload.
See:
http://linux.die.net/man/3/setsockopt
The options are SO_SNDBUF, and SO_RCVBUF. If you directly use the C-API, the call is setsockopt itself. If you use some kind of framework look up how to set socket options. This is indeed a kernel-side buffer, not one held by your process. It determines how many bytes the kernel can hold ready for you to fetch from a call to read/receive. It also affects the flow control mechanism of TCP.
You are being told to increase the socket send or receive buffer sizes. These are associated with the socket, in the TCP part of the kernel. See setsockopt() and SO_RCVBUF and SO_SNDBUF.
I am working on analyzing H264 video data being streamed over a network. Right now, I am able to successfully extract and analyze the raw H264 for UDP. This process is going to be ALOT harder for the TCP/RTSP because of fragmentation and multiplexing.
Is the video compression / encoding any different on the TCP/RTSP multiplexed stream compared to the UDP stream?
It's only slightly harder as you typically have to demultiplex the audio and video, as well as the RTCP reports on the TCP connection. Fragmentation is not an issue.
Is the video compression / encoding any different on the TCP/RTSP multiplexed stream compared to the UDP stream?
No differences at all. The multiplexing of RTP/RTCP packets is defined in RFC2326.
As far as tools go, you can use openRTSP from http://www.live555.com which handles the transport for you (RTP over RTSP via the -t command line argument) and writes the frames to file.
With reference to Ainitak's comment, it's not that complex: there's a 4 byte header, '$' followed by the channel id, followed by the 2 byte length of the following RTP/RTCP packet. It's not too tricky to parse this.
My question is that when a socket at the receiver-side sends an ack? At the time the application read the socket data or when the underlying layers get the data and put it in the buffer?
I want this because I want both side applications know whether the other side took the packet or not.
It's up to the operating system TCP stack when this happens, since TCP provides a stream to the application there's no guarenteed 1:1 correlation between the application doing read/writes and the packets sent on the wire and the TCP acks.
If you need to be assured the other side have received/processed your data, you need to build that into your application protocol - e.g. send a reply stating the data was received.
TCP ACKs are meant to acknowledge the TCP packets on the transmission layer not the application layer. Only your application can signal explicitly that it also has processed the data from the buffers.
TCP/IP (and therefor java sockets) will guarantee that you either successfully send the data OR get an error (exception in the case of java) eventually.
This might be a silly question:
Does HTTP ever use the User Datagram Protocol?
For example:
If one is streaming MP3 or video using HTTP, does it internally use UDP for transport?
From RFC 2616:
HTTP communication usually takes place
over TCP/IP connections. The
default port is TCP 80, but other
ports can be used. This does not
preclude HTTP from being implemented
on top of any other protocol on the
Internet, or on other networks. HTTP
only presumes a reliable transport;
any protocol that provides such
guarantees can be used; the mapping
of the HTTP/1.1 request and response
structures onto the transport data
units of the protocol in question is
outside the scope of this
specification.
So although it doesn't explicitly say so, UDP is not used because it is not a "reliable transport".
EDIT - more recently, the QUIC protocol (which is more strictly a pseudo-transport or a session layer protocol) does use UDP for carrying HTTP/2.0 traffic and much of Google's traffic already uses this protocol. It's currently progressing towards standardisation as HTTP/3.
Typically, no.
Streaming is seldom used over HTTP itself, and HTTP is seldom run over UDP. See, however, RTP.
For something as your example (in the comment), you're not showing a protocol for the resource. If that protocol were to be HTTP, then I wouldn't call the access "streaming"; even if it in some sense of the word is since it's sending a (possibly large) resource serially over a network. Typically, the resource will be saved to local disk before being played back, so the network transfer is not what's usually meant by "streaming".
As commenters have pointed out, though, it's certainly possible to really stream over HTTP, and that's done by some.
Maybe just a bit of trivia, but UPnP will use HTTP formatted messages over UDP for device discovery.
Yes, HTTP, as an application protocol, can be transferred over UDP transport protocol.
Here are some of the services that use UDP and an underlying protocol for transferring HTTP data and streaming it to the end-user:
XMPP's Jingle Raw UDP Transport Method
A number for services that use UDT --- UDP-based Data Transfer Protocol, which is the a superset of UDP protocol.
The Transport Layer Security (TLS) protocol encapsulating HTTP as well as the above mentioned XMPP and other application protocols does have an implementation that uses UDP in its transport layer; this implementation is called Datagram Transport Layer Security (DTLS).
Push notifications in GNUTella are HTTP requests sent over UDP transport.
This article contains further details on streaming over UDP and its reliable superset, the RUDP: Reliable UDP (RUDP): The Next Big Streaming Protocol?
Of course, it doesn't necessarily have to be transmitted over TCP. I implemented HTTP on top of UDP, for use in the Satellite TV Broadcasting industry.
If you are streaming an mp3 or video that may not necessarily be over HTTP, in fact I'd be suprised if it was. It would probably be another protocol over TCP but I see no reason why you cannot stream over UDP.
If you do you have to take into account that there is no certainty that your data will arrive at the other end, but I can take it that you know about UDP.
To answer you question, No, HTTP does NOT use UDP.
For what you talk about though, mp3/video streaming COULD happen over UDP and in my opinion should never happen over HTTP.
Maybe some change on this topic with QUIC
QUIC (Quick UDP Internet Connections, pronounced quick) is an experimental transport layer network protocol developed by Google and implemented in 2013. QUIC supports a set of multiplexed connections between two endpoints over User Datagram Protocol (UDP), and was designed to provide security protection equivalent to TLS/SSL, along with reduced connection and transport latency, and bandwidth estimation in each direction to avoid congestion. QUIC's main goal is to optimize connection-oriented web applications currently using TCP.
I think some of the answers are missing an important point. The choice between UDP and TCP should not be based on the type of data (e.g., audio or video) or whether the application starts to play it before the transfer is completed ("streaming"), but whether it is real time. Real time data is (by definition) delay-sensitive, so it is often best sent over RTP/UDP (Real Time Protocol over UDP).
Delay is not an issue with stored data from a file, even if it's audio and/or video, so it is probably best sent over TCP so any packet losses can be corrected. The sender can read ahead and keep the network pipe full and the receiver can also use lots of playout buffering so it won't be interrupted by the occasional TCP retransmission or momentary network slowdown. The limiting case is where the entire recording is transferred before playback begins. This eliminates any risk of a playback stall, but is often impractical.
The problem with TCP for real-time data isn't retransmissions so much as excessive buffering as TCP tries to use the pipe as efficiently as possible without regard to latency. UDP preserves application packet boundaries and has no internal storage, so it does not introduce any latency.
(This is an old question, but it deserves an updated answer.)
In all likelihood, HTTP/3 will be using the QUIC protocol, which is described as
multiplexed transport over UDP
So, from a certain point of view, you could say that HTTP/3 will be using UDP.
The answer: Yes
Reason: See the OSI model.
Explaination:
HTTP is an application layer protocol, which could be encapsulated with a protocol that uses UDP, providing arguably faster reliable communication than TCP. The server daemon and client would obviously need to support this new protocol. Quake 2 protocol proves that UDP can be used over TCP to provide a basis for a structured communication system insuring flow control (e.g. chunk ids).
http over udp is used by some torrent tracker implementations (and supporteb by all main clients)
In theory yes it is possible to use UDP for http but that might be problematic. Say for instance in your example a mp3 or a video is being streamed there will be problem of ordering and some bits might go missing as UDP is not connection oriented there is no retransmit mechanism.
HTTP/3 (aka QUIC) uses UDP instead of TCP.
https://http3-explained.haxx.se/en/the-protocol/feature-udp
UDP is the best protocol for streaming, because it doesn't make demands for missing packages like TCP. And if it doesn't make demands, the flow is far more faster and without any buffering.
Even the stream delay is lesser than TCP. That is because TCP (as a far more secure protocol) makes demands for missing packages, overwriting the existing ones.
So TCP is a protocol too advanced to be used for streaming.