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.
I'm working on a server/client socket application that is using Linux TUN interface.
Server gets packets directly from TUN interface and pass them to clients and clients put received packets directly in the TUN interface.
<Server_TUN---><---Server---><---Clients---><---Client_TUN--->
Sometimes the packets from Server_TUN need to be fragmented in IP layer before transmitting to a client.
So at the server I read a packet from TUN, start fragmenting it in the IP layer and send them via socket to clients.
When the fragmentation logic was implemented, the solution did not work well.
After starting Wireshark on Client_TUN I noticed for all incoming fragmented packets I get TCP Checksum error.
At the given screenshot, frame number 154 is claimed to be reassembled in in 155.
But TCP checksum is claimed to be incorrect!
At server side, I keep tcp data intact and for the given example, while you see the reverse in Wireshark, I've split a packet with 1452 bytes (including IP header) and 30 bytes (Including IP header)
I've also checked the TCP checksum value at the server and its exactly is 0x935e and while I did not think that Checksum offloading matters for incoming packets, I checked offloading at the client and it was off.
$ sudo ethtool -k tun0 | grep ": on"
scatter-gather: on
tx-scatter-gather: on
tx-scatter-gather-fraglist: on
generic-segmentation-offload: on
generic-receive-offload: on
tx-vlan-offload: on
tx-vlan-stag-hw-insert: on
Despite that, because of the solution is not working now, I don't think its caused by offload effect.
Do you have any idea why TCP checksum could be incorrect for fragmented packets?
Hopefully I found the issue. It was my mistake. Some tcp data was missing when I was coping buffers. I was tracing on the indexes and lengths but because of the changes in data, checksum value was calculating differently in the client side.
In TCP/IP, we have MSS and MTU when sending and receiving packets.
MTU is an IP layer concept, which is determined by the underlying hardware. It shows the maximum data size that an IP layer packet can contain during one transmission.
MSS is a TCP layer concept, which is limited by the MTU, showing that the TCP data stream will be fragmented into MSS-size packets.
Our protocol lies on top of TCP, and each protocol will define its own packet. One example is MySQL, which defines its packet size up to 2^24-1, that is around 16M. When the big enough protocol packet comes to TCP, it will be fragmented according to MSS.
Assume that a client needs to send DATA1 and DATA2 to server. DATA2 size is bigger than MSS, and DATA2 will be fragmented into DATA2_1, DATA2_2. As the packets will be handled by the IP layer, so the time that each packet arrives at server might not be the same as that when the client sends them.
So I think the sequence of packets' arriving might be the following:
DATA1 DATA2_1, DATA2_2
DATA1, DATA2_1, DATA2_2
DATA1, DATA2_2, DATA2_1
In the first case, the server receives DATA1 and DATA2_1 in one tcp packet and then another packet contains DATA2_2 arrives.
In the second case, the server receives DATA1, DATA2_1 and DATA2_2 in three packets.
In the third case, the server first receives DATA2_2 and then DATA2_1.
My question:
Is the third case possible?
If yes, it disobeys that TCP is a stream protocol, and stream protocol should be ordered. And even this does not break the stream rule, how to handle this scenario?
If no, how TCP makes the disordered packets into its original order?
It is possible to receive that sequence over the network, however the TCP implementation will hide that detail from your application and only feed the data to you in stream order. (In fact since fragmentation happens at the IP layer it won't even be shown to the TCP layer until the second part has arrived also)
The fact that received packets have to be held in a buffer even after receiving them in some cases like this is why you will see people referring to UDP as better for lower latency applications: you can receive datagrams out of order with UDP and it's up to your application to figure out how to deal with that possibility.
Is the third case possible.
Yes, of course.
If so, it disobeys that TCP is a stream protocol ...
No it doesn't.
Your cases concern arrival of IP packets into a host. TCP being a stream protocol is about delivery of data into an application.
The packet fragments get reassembled in the correct order by the IP layer, and the packets get reassembled into segments in the correct order by TCP, and the now correctly ordered data stream is delivered to the application.
I understand that in the three-way handshake, sometimes the receiving end will send a SYNACK packet when establishing a connection (piggybacking), but when would it ever send a SYN and then an ACK packet?
For example:
->SYN
<-SYN_ACK
->ACK
versus:
->SYN
<-SYN
->SYN_ACK
Thanks!
No it won't - here's the reason why
SYN is typically sent by the 'client' (eg. your browser) when it wants to open a TCP connection to a server (eg. your web server). A server has no way of 'knowing' beforehand which client wants to open a connection (and hence send a SYN) to it. So it cannot send an unsolicited SYN.
SYN and ACK are flags, so the SYN-ACK from server is an ACK to the client's SYN (and it's own SYN). Technically, it can send them separately, but, sending SYN and ACK separately would involve additional half round trip. 'cos then it would be a four way handshake ((c)SYN -> , <- SYN(s), <-ACK(s), (c)ACK ->) that doesn't achieve any more reliability than three way handshake offers. Consequently it makes no sense to do that way.
Having said so you could theoretically design a protocol with 4 way handshake, but TCP isn't designed so.
Hope that helps.
Why is the IP called a connectionless protocol? If so, what is the connection-oriented protocol then?
Thanks.
Update - 1 - 20:21 2010/12/26
I think, to better answer my question, it would be better to explain what "connection" actually means, both physically and logically.
Update - 2 - 9:59 AM 2/1/2013
Based on all the answers below, I come to the feeling that the 'connection' mentioned here should be considered as a set of actions/arrangements/disciplines. Thus it's more an abstract concept rather than a concrete object.
Update - 3 - 11:35 AM 6/18/2015
Here's a more physical explanation:
IP protocol is connectionless in that all packets in IP network are routed independently, they may not necessarily go through the same route, while in a virtual circuit network which is connection oriented, all packets go through the same route. This single route is what 'virtual circuit' means.
With connection, because there's only 1 route, all data packets will arrive in the same order as they are sent out.
Without connection, it is not guaranteed all data packets will arrive
in the same order as they are sent out.
Update - 4 - 9:55 AM 2016/1/20/Wed
One of the characteristics of connection-oriented is that the packet order is preserved. TCP use a sequence number to achieve that but IP has no such facility. Thus TCP is connection-oriented while IP is connection-less.
The basic idea is pretty simple: with IP (on its own -- no TCP, UDP, etc.) you're just sending a packet of data. You simply send some data onto the net with a destination address, but that's it. By itself, IP gives:
no assurance that it'll be delivered
no way to find out if it was
nothing to let the destination know to expect a packet
much of anything else
All it does is specify a minimal packet format so you can get some data from one point to another (e.g., routers know the packet format, so they can look at the destination and send the packet on its next hop).
TCP is connection oriented. Establishing a connection means that at the beginning of a TCP conversation, it does a "three way handshake" so (in particular) the destination knows that a connection with the source has been established. It keeps track of that address internally, so it can/will/does expect more packets from it, and be able to send replies to (for example) acknowledge each packet it receives. The source and destination also cooperate to serial number all the packets for the acknowledgment scheme, so each end knows whether packets it sent were received at the other end. This doesn't involve much physically, but logically it involves allocating some memory on both ends. That includes memory for metadata like the next packet serial number to use, as well as payload data for possible re-transmission until the other side acknowledges receipt of that packet.
TCP/IP means "TCP over IP".
TCP
--
IP
TCP provides the "connection-oriented" logic, ordering and control
IP provides getting packets from A to B however it can: "connectionless"
Notes:
UDP is connection less but at the same level as TCP
Other protocols such as ICMP (used by ping) can run over IP but have nothing to do with TCP
Edit:
"connection-oriented" mean established end to end connection. For example, you pick up the telephone, call someone = you have a connection.
"connection-less" means "send it, see what happens". For example, sending a letter via snail mail.a
So IP gets your packets from A to B, maybe, in any order, not always eventually. TCP sorts them out, acknowledges them, requests a resends and provides the "connection"
Connectionless means that no effort is made to set up a dedicated end-to-end connection, While Connection-Oriented means that when devices communicate, they perform handshaking to set up an end-to-end connection.
IP is an example of the Connectionless protocols , in this kind of protocols you usually send informations in one direction, from source to destination without checking to see if the destination is still there, or if it is prepared to receive the information . Connectionless protocols (Like IP and UDP) are used for example with the Video Conferencing when you don't care if some packets are lost , while you have to use a Connection-Oriented protocol (Like TCP) when you send a File because you want to insure that all the packets are sent successfully (actually we use FTP to transfer Files). Edit :
In telecommunication and computing in
general, a connection is the
successful completion of necessary
arrangements so that two or more
parties (for example, people or
programs) can communicate at a long
distance. In this usage, the term has
a strong physical (hardware)
connotation although logical
(software) elements are usually
involved as well.
The physical connection is layer 1 of
the OSI model, and is the medium
through which the data is transfered.
i.e., cables
The logical connection is layer 3 of
the OSI model, and is the network
portion. Using the Internetwork
Protocol (IP), each host is assigned a
32 bit IP address. e.g. 192.168.1.1
TCP is the connection part of TCP/IP. IP's the addressing.
Or, as an analogy, IP is the address written on the envelope, TCP is the postal system which uses the address as part of the work of getting the envelope from point A to point B.
When two hosts want to communicate using connection oriented protocol, one of them must first initiate a connection and the other must accept it. Logically a connection is made between a port in one host and other port in the other host. Software in one host must perform a connect socket operation, and the other must perform an accept socket operation. Physically the initiator host sends a SYN packet, which contains all four connection identifying numbers (source IP, source port, destination IP, destination port). The other receives it and sends SYN-ACK, the initiator sends an ACK, then the connection are established. After the connection established, then the data could be transferred, in both directions.
In the other hand, connectionless protocol means that we don't need to establish connection to send data. It means the first packet being sent from one host to another could contain data payloads. Of course for upper layer protocols such as UDP, the recipient must be ready first, (e.g.) it must perform a listen udp socket operation.
The connectionless IP became foundation for TCP in the layer above
In TCP, at minimal 2x round trip times are required to send just one packet of data. That is : a->b for SYN, b->a for SYN-ACK, a->b for ACK with DATA, b->a for ACK. For flow rate control, Nagle's algorithm is applied here.
In UDP, only 0.5 round trip times are required : a->b with DATA. But be prepared that some packets could be silently lost and there is no flow control being done. Packets could be sent in the rate that are larger than the capability of the receiving system.
In my knowledge, every layer makes a fool of the one above it. The TCP gets an HTTP message from the Application layer and breaks it into packets. Lets call them data packets. The IP gets these packets one by one from TCP and throws it towards the destination; also, it collects an incoming packet and delivers it to TCP. Now, TCP after sending a packet, waits for an acknowledgement packet from the other side. If it comes, it says the above layer, hey, I have established a connection and now we can communicate! The whole communication process goes on between the TCP layers on both the sides sending and receiving different types of packets with each other (such as data packet, acknowledgement packet, synchronization packet , blah blah packet). It uses other tricks (all packet sending) to ensure the actual data packets to be delivered in ordered as they were broken and assembled. After assembling, it transfers them to the above application layer. That fool thinks that it has got an HTTP message in an established connection but in reality, just packets are being transferred.
I just came across this question today. It was bouncing around in my head all day and didn't make any sense. IP doesn't handle transport. Why would anyone even think of IP as connectionless or connection oriented? It is technically connectionless because it offers no reliability, no guaranteed delivery. But so is my toaster. My toaster offers no guaranteed delivery, so why not call aa toaster connectionless too?
In the end, I found out it's just some stupid title that someone somewhere attached to IP and it stuck, and now everyone calls IP connectionless and has no good reason for it.
Calling IP connectionless implies there is another layer 3 protocol that is connection oriented, but as far as I know, there isn't and it is just plain stupid to specify that IP is connectionless. MAC is connectionless. LLC is connectionless. But that is useless, technically correct info.