It is a common knowledge that transferring TCP packet inside a tunnel with TCP connection can create a devastating effect called TCP meltdown and degrade tunnel quality greatly. I somehow wondering if similar effect may happen in we try to transfer TCP data over a Quic connection. Even though Quic is UDP packets, but it need to have something similar to windowing for keeping track of received packets in order to provide a connection-oriented protocol. So I'm not sure if a similar effect will happen or not.
Any idea?
QUIC indeed uses a similar congestion control as TCP, see https://www.rfc-editor.org/rfc/rfc9002.html#name-congestion-control. So when tunnelling a TCP connection over a QUIC stream, I would say the same "meltdown" problems can occur (a QUIC stream has the same properties as a TCP connection: reliable ordered stream of byte, so the stream will stall if QUIC packets are lost).
However, a QUIC extensions is being defined for sending datagrams, https://datatracker.ietf.org/doc/html/draft-ietf-quic-datagram. That might provide a better way for transporting TCP packets, as these datagrams will never be retransmitted at QUIC level. However, it would require TCP packets to fit into a Datagram frame.
Related
One of the key differences between HTTP/2 and HTTP/3 is the switch from TCP to UDP.
As I understand it, TCP verifies data integrity by verifying that no data packets have been lost. Any packets lost are requested again to ensure all data is correctly received.
For UDP, there is no such validation. If packets are lost then so be it.
That being said, if I make a request on HTTP/3 and a packet is lost, is there a mechanism to ensure I get all my data, or will there be a risk that my response will be missing data packets?
If packets are lost then so be it.
No, with UDP it is no "so be it" but it is up to the protocol on top of UDP to care about packet loss, duplication and reordering or not to care. For example with RTP (real time audio in VoIP etc) it is fine to have some packet loss, since there is no use for later arriving packets anyway (audio must be low latency). And reordering and duplication is handled in RTP with protocol-inherent sequence numbers.
For HTTP/3 instead data loss is not acceptable. HTTP/3 is build on top of QUIC which is build on top of UDP. Packet loss is handled within QUIC (see QUIC Loss Detection and Congestion Control). Thus HTTP/3 is basically build on top of a reliable transport (QUIC) the same as HTTP/1 and HTTP/2 are build on top of an reliable transport layer (TCP).
TCP has a greater computation overhead to ensure reliable delivery of packets. But, since modern networks are fast, is there any scenario in which performance of UDP outweighs the reliability of TCP?
Is there any other particular advantage of UDP over TCP?
I can see two cases, where UDP would have an upper hand over TCP.
First, one of the attractive features of UDP is that since it does not need to retransmit lost packets nor does it do any connection setup, sending data incurs less delay. This lower delay makes UDP an appealing choice for delay-sensitive applications like audio and video.
Second, multicast applications are built on top of UDP since they have to do point to multipoint. Using TCP for multicast applications would be hard since now the sender would have to keep track of retransmissions/sending rate for multiple receivers.
It depends on your usage. If your application is time sensitive, like Voice over IP, then you don't care about missing packets. What you care about is the delay in that case.
You should have a look at this answer: What are examples of TCP and UDP in real life?
You could also look at the Wikipedia related section: http://en.wikipedia.org/wiki/User_Datagram_Protocol#Comparison_of_UDP_and_TCP
Applications that require constant data flow, bulk data and which require fastness than reliability uses UDP over TCP.
udp provides better application level control over what data is sent....since the data is packaged in a udp segment and immediately passed over to the network layer......hence no-frills segment delivery service is observed.
There is no need for connection establishment hence no delay(unlike tcp...which requires handshaking before the actual data transfer)
There is no need to maintain connection state in the end systems(ie no need for send and receive buffers,congestion control parameters and sequence and acknowledgement number parameters)..hence more active clients could be supported
Small packet header overhead for udp(only 8 bytes) where as tcp has 20 bytes of header
Facebook uses UDP connections instead of TCP/IP to connect to theirs Memcached Servers
There are couple of differences of UDP over TCP.
First, TCP is connection-based whereas UDP is connectionless.
Connection-based: Make sure that all messages will arrive and arrive in the correct order.
Connectionless: It does not guarantee order or completeness.
Second, Here is why UDP is faster over TCP:
UDP does not require ACK message back
UDP has no flow control
No duplication verification at the receiving end
Shorter header
I had a strange idea. I heard of software which from my understanding uses UDP to transfer files decreasing the overhead found in TCP packets.
If my app requires TCP and my LAN has software set up to communicate with another datacenter on the other side of the coast with software setup on their end. Would it be possible to send the actual data via UDP but than simulating TCP on both ends?
Does anyone have any ideas or information about such projects?
If you're asking if you can use UDP as a Layer 2, then the answer is yes, sort of. There are various protocols that allow you to create a tunnel to another network using a UDP transport, such as L2TP and even IPsec (with NAT traversal). You could also do it at the application layer.
If you're asking if TCP can be implemented in UDP, the answer is no. First, TCP packets and UDP packets have an incompatible format. Second, TCP and UDP have different protocol numbers (seen in the IP header) which means that TCP traffic destined for a UDP port would not be passed to the correct upper-layer protocol.
Both TCP and UDP are built on top of the IP, but the TCP uses different packet structure and at the layer-2 it is not possible to mimic the TCP using UDP packets.
Of course, if you have the control on both the source and destination, then it is possible to create a reliable UDP tunnel for the TCP packets. This would require some internal information (packet number, ack/nack flags) in the body of the UDP packet.
There is an interesting project http://udt.sourceforge.net/
It is a broadcast-capable reliable file transfer mechanism built on top the UDP.
PseudoTCP is a protocol which implements TCP algorithms on top of the UDP. It was introduced since the NAT traversal for TCP is much more complicated than UDP. But some P2P applications do need a reliable data transfer among nodes.
So far as I know, there are two PseudoTCP variations: Libjingle and Libnice.Libjingle is an open source library from google which was initially for gtalk. You could take a look at file sharing example from libjingle: https://developers.google.com/talk/libjingle/file_share. Recently, Chrome desktop also use PseudoTCP implementation from libjingle for reliable connections.
Yes, you can develop a protocol on UDP that simulates TCP. However, if you simulated TCP fully, it would technically have more overhead. Because TCP is implement as the packet and your simulated TCP is implemented in the body of the packet.
If you only need one or two features of TCP (such as basic ordering), then implementing it in UDP is useful.
Halo uses 2-3 (IIRC) UDP protocols that simulate different features of TCP, then full fledged TCP for initializing game-states. I Shot You First Networking, GDC publication
For example, in one case, they send 3 duplicate UDP packets to overcome packet loss.
If you control the software on both ends, and it is cost-effective to build your own protocol, then UDP can be versatile.
One way to do it now on Linux-3.18+ is to use Foo over UDP (FOU) which implements Generic UDP Encapsulation (GUE). Here's a good introduction to FOU, and the man page for ip-fou.
Or if you want an [open source] UDP based file transfer system there are things like UDT, UFTP, Tsunami-UDP, and even Google's QUIC (Now deprecated in favour of IETF QUIC).
Update: The QUIC protocol now has been standardised by the IETF which provides for secure reliable and unreliable transport over UDP as an alternative to TCP. There's a wide range of QUIC implementations available. There is also a growing set of protocol mappings on to QUIC such as HTTP/3, DNS over QUIC, etc
If my app requires TCP and my LAN has software setup to communicate
with another datacenter on the other side of the coast with software
setup on their end. Would it be possible to send the actual data via
UDP but than simulating TCP on both ends?
No. A UDP socket is in a different namespace from a TCP socket. You will be unable to write UDP at one end and send or receive TCP at the other end. TCP and UDP are peer protocols; both exist at the layer above IP. You can't use one to spoof the other.
Hmm, I believe so. You'd need to use a proxy at both ends, but it should be possible.
The biggest problem you are going to run into is that UDP is designed with the idea that you don't care if some of the packets don't ever make it to the other end.
Here's a link with some more info:
http://www.cyberciti.biz/faq/key-differences-between-tcp-and-udp-protocols/
IMHO, it's not a good idea to transmit files via UDP.
TCP's problems are in its algorithms, not its headers.
You certainly could implement the TCP algorithms on top of UDP. That would effectively be the same as tunneling TCP datagrams inside of UDP datagrams. But all this accomplishes is to add a few more bytes of overhead to each packet, and require another endpoint to unwrap the packets.
UDP itself is just thin shim on top of IP: its a convenient way to access IP packet switched networking without having to dive into kernels or receive special handling from routers. The main reason to implement reliable transport on top of UDP is to get away from TCP algorithms in favor of something more efficient. FileCatalyst was mentioned above as one company which does this, and my own company Data Expedition, Inc. does so as well.
So you could implement TCP algorithms on top of UDP, but you wouldn't want to.
You can simulate something like a connection over UDP, and you as well can add reliability checks and ordering and retransmission and so on. - but then, it still isn't TCP, it just acts the way.
Of course, one of the ends can be a kind of "hub" or "proxy" which does an adaption. Then you don't have a 2-end solution, but in fact a 4 end solution - one pair with "real" TCP and the other with the "self-knitted" "TCP" - which you put together with an appropriately crafted program.
If a TCP connection is established between a client and server, is sending data faster on this connection-oriented route compared to a connectionless given there is less header info in the packets? So a TCP connection is opened and bytes sent down the open connection as and when required. Or would UDP still be a better choice via a connectionless route where each packet contains the destination address?
Is sending packets via an established TCP connection (after all hand shaking has been done) a method to be faster than UDP?
I suggest you read a little bit more about this topic.
just as a quick answer. TCP makes sure that all the packages are delivered. So if one is dropped for whatever reason. The sender will continue sending it until the receiver gets it. However, UDP sends a packet and just forgets it, so you might loose some of the packets. As a result of this, UDP sends less number of packets over network.
This is why they use UDP for videos because first loosing small amount of data is not a big deal plus even if the sender sends it again it is too late for receiver to use it, so UDP is better. In contrast, you don't want your online banking be over UDP!
Edit: Remember, the speed of sending packets for both UDP and TCP is almost same, and depends on the network! However, after handshake is done in TCP, still receiver needs to send the acks, and sender has to wait for ack before sending new batch of data, so still would be a little slower.
In general, TCP is marginally slower, despite less header info, because the packets must arrive in order, and, in fact, must arrive. In a UDP situation, there is no checking of either.
I was told that for a High Frequency Trading (HFT) system that requires low-latency, TCP is used over UDP. I was told that with TCP you can make point to point connections, whereas you cannot with UDP, however from my understanding you can send UDP packets to specific IP/port.
There are several arguments used in this article as to why UDP > TCP for gaming but I can see relevance for HFT.
Why would TCP be a better protocol to use for HFT?
(Admins: My previous post of this question was silently removed with no explanation. If I am violating terms of use please alert me of this instead of silently removing the question)
UDP is superior to TCP if you don't need some of the features TCP provides. Every feature has a cost, and so if you don't need features, you are paying that cost for no reason.
In an HFT application, you need pretty much every feature TCP requires. So if you picked UDP, you'd have to implement those features yourself. That means you'd have to implement connection establishment, connection teardown, retransmissions, transmit pacing, windows, and so on.
If there was a way to do all those things that was better than the way TCP was doing it, TCP would be doing it that way. You'd have one hand tied behind your back because TCP is heavily optimized by some of the best minds on the planet and implemented in/with the kernel.
There's no reasons to expect a stream of data over an already-established TCP connection would be slower than the same data over UDP, plus you get checksumming, retries, and all the other TCP goodness. UDP mainly wins in cases where you can afford to discard the reliability or where the overhead of many TCP handshakes would be too expensive, such as with common DNS queries.
TCP is faster for when using a few connections, the important difference is that modern NICs perform significant amounts of acceleration on TCP and not really that much for UDP. This means there is more overhead to process each UDP packet and as such they cannot compete unless you need to send to multiple recipients simultaneously.
However the UDP multicast route still suffers the same problems as unicast UDP per datagram overheads. Therefore many HFT systems use hardware accelerated systems that can multiplex the streams across many NICs via TCP, example Solace.
These days though you want to completely bypass the kernel with say a userspace IP stack such as by Solarflare or Mellanox, or even skip both the kernel and IP stack with RDMA.
Quite simply, if you need connection reliability (ensuring that every byte of data transmitted is received), you should be using TCP regardless.
As you mentioned, UDP is more suitable for games, where 100% accurate real-time tracking of every object would use quite a large amount of bandwidth and is unnecessary (this is where slow connections encounter lag).
There is no special difference between a TCP port and a UDP port, beyond the type of connection being used (send the packet and forget it, UDP style, or negotiate a connection and sustain it, TCP style) and the service listening on the server side. e.g. TCP/25 would usually reveal a SMTP server, whereas UDP/25 would not.
Basically, modern TCP implementations are going to be just as fast as UDP, if you're keeping the connection alive. If TCP is having to resend a packet, you'd need to resend it in UDP too. Plus for UDP you're going to end up implementing the same reliability code (retransmission of dropped packets) that TCP has already implemented.