I require to reduce latency of windows application using RDMA,but I have control over only one side of communication,RDMA(Kernel Bypass) implementation irrespective of other side.
Is it possible to implement RDMA (RoCE) only on client side without worrying about Server RDMA support?
No, RDMA (RoCE, iWARP, whatever) is a wire protocol, so both sides need to handle it. An analogy would be using compression or pipelining for HTTP - you need support on both the client and the server.
However, on the other hand, you don't need a hardware implementation of RoCE on both sides. You could use a true RDMA adapter that offloads data movement to hardware on the client side, and use a software implementation of RoCE on the server side (for example, Soft RoCE).
No, you cannot.
RDMA require both machines to handle the connection setup, and communication. And it requires the hardware and driver of NICs and configurations of both sides and switches of the network to support RDMA protocols.
No, it's not. As at most of the transport protocols that's not possible. (e.g. TCP, UDP)
- As additional suggestion: If you just have one adapter with RDMA support you could use on the other side Soft-RoCE
-
Related
I want to make sure, if use UDP within a host, should i care about the package lost issue?
Yes, you should care about reliability when using UDP. Even if you use it on localhost, there is no guaranty that packets are not lost because the Protocol Specifications does not ensure this. It also depends on the implementation of UDP in Operating System. It may behave differently on different operating systems as far as reliability is concerned because there is no rule defined in UDP specifications.
Also the order of delivery in UDP is not ensured so you should also take care of it while using UDP for IPC.
I hope it helps.
What is the purpose for using sip with VOIP ? is it just to know the remote ip address ?
If i know (by server) the remote IP address ,then established a direct TCP socket connection for call negotiation, and send the media over RTP protocol ,so am i still need sip protocol ? or how can sip help me here?
The Session Initiation Protocol does rather a lot more than find out a remote IP address/port/transport triple.
It lets two parties
negotiate the media streams (including codecs and transports) and
establish commonly understood extensions to the protocol.
It also describes how to build scalable infrastructure (proxies, using SRV and NAPTR records, back to back user agents), location services and a host of other details that go into making a voice (or any other kind of) call to arbitrary third parties.
then you have implemented a sip alternative. sip (session instanciation protocol) does just control the "phone call". if you want to do that on your own, why not? the only problem would be that there are many sip clients and just one (or few) clients using your protocol.
SIP allocates a IP:port to a voip call. The RTP flows (one for each direction) will then use this IP:port as a destination address. If you have only one static RTP flow to send to your server, it may be useful and ok to do what you said.
Otherwise, if there are many clients, or if your system has to change a lot, it's better to use a polished protocol which will dynamically allocate ports and establish your sessions.
nobody forces you to implement a standard.
e.g. why do you implement the media stream in RTP? most likely because you already have code that "talks" RTP (e.g. a library, or a raedymade application).
the nice thing about standards is, that it will work "out-of-the-box" with all other applications implementing the same standard.
if it is an open standard, there's another nice thing: other people have already spent a lot of brain power into getting the implementation right. you don't need to fall into the same problems.
It’s just a start where I am exploring more in P2P side, and finding reasons in terms of Scalability or anything else for : SIP or XMPP (Jingle) for following use case :
P2P Client Application Capable to perform File Transfer on all Network Traversal Scenarios.
// For Signaling (e.g.; to connect/locate/disconnect peers) both XMPP (Jingle) or SIP are available.
May I know possible reasons to use what and why? Any practical use? e.g.; Scalability or anything which really makes a difference for the above Use Case
Jingle is an XMPP extension to handle multimedia sessions. In effect Jingle is the XMPP equivalent of SIP.
As far as a P2P file application goes:
Jingle and SIP are roughly equivalent as far as scalability goes. Both separate the signalling and media providing more flexibility (and consequently complications) with the way server side components can be deployed.
XMPP/Jingle has a better security design making it much more practicable to enforce clients using an SSL signalling layer. SIP does support SSL but it's more convoluted and also doesn't enjoy widespread support in the real World,
As far as NAT goes you're going to have the same problems with both. The scalability you get from having separate signalling and media paths comes back to bite when NAT is involved. There are a few different mechanisms to deal with NAT the latest attempt is ICE. ICE is collection of different mechanisms to try and resolve different NAT configurations and it's worth bearing in mind that not all configurations can be resolved and the fallback is to use a media proxying server such as TURN.
If I was you I'd use XMPP but before starting I'd work out exactly what NAT configurations need to be supported. If you need to support arbitrary clients from anywhere on the internet then you will not be able to rely on always being able to establish direct P2P communications between your clients and that's where you will face your biggest challenge.
I have to implement IPC mechanism (Sending short messages) between java/c++/python process running on the same system. One way to implement is using socket using TCP protocol. This requires maintain connection and other associated activities.
Instead I am thinking of using UDP protocol which does not requires connection and I can send messages.
My question is , does UDP on same machine ( for IPC ) still has same disadvantage has it is applicable when communicating across machines ( like un reliable packet delivery, out of order packet.
Yes, is still unrealiable. For local communication try to use named pipes or shared memory
Edit:
Don't know the requirements of your applications, did you considered something like MPI (altough Java is not well supported...) or, Thrift? ( http://thrift.apache.org/ )
Local UDP is still unreliable, but the major advantage is UDP multicast. You can have one data publisher and many data subscribers. The kernel does the job of delivering a copy of the datagram to each subscriber for you.
Unix local datagram sockets, on the other hand, are required to be reliable but they do not support multicast.
Local UDP is more unreliable than on a network, like 50%+ packet drop unreliable. It is a terrible choice, kernel developers have attributed the quality down to lack of demand.
I would recommend investigating message based middleware preferably with a BSD socket compatible interface for easy learning curve. A suggestion would be ZeroMQ which includes C++, Java and Python bindings.
Local UDP is both still unreliable and sometimes blocked by firewalls. We faced this in our MsgConnect product which uses local UDP for interthread communication. BTW MsgConnect can be an option for your task so that you don't need to deal with sockets. Unfortunately there's no Python binding, but "native" C++ and Java implementations exist.
Does anyone know why should I use Winpcap and not just .Net sockets to sniff packets on my local pc?
TY
Sockets (.NET, Winsock, etc.) normally collect at layer 7, the Application layer. That is, whatever is sent by the sender is what is received by the receiver. All of the various headers that are added automatically on the sending side are stripped off by the time the receiver reads the data from the socket.
It is possible to configure a socket to be a raw socket, in which case, you can see all of the headers down to layer 3, the Network layer. Further still, you can put the raw socket in promiscuous mode, which allows you to see all of the traffic on the network, not just the packets destined for your machine. But even this is limited. For example, when you configure the raw socket, you specify the protocol type to use, e.g., IP, ICMP, etc. This limits the socket to "seeing" packets that adhere to that protocol. I have been unable to figure out how to make the socket see all packets at layer 3 regardless of protocol.
Winpcap operates as a device driver at layer 2, the Data Link layer. In this case, you see literally all of the packets on the network with full headers down to layer 2. Winpcap also offers filtering capability so you can narrow down the packets that are reported to you based on whatever criteria you provide.
As far as choosing between them, it really boils down to the requirements of your specific task. If you are trying to implement any kind of realistic network analysis capability, you'll be hardpressed to do that with just sockets. Winpcap makes more sense in that case. However, if you are only interested in IP packets, for example, then sockets will work fine for that.
As far as I understanf .Net sockets are an IPC to communicate between 2 processes. While winpcap is a library that help you to access the data link layer an sniff pacquets going through your network hardware (or virtual) devices on your machine. Data link layer allow to get the data on any socket (.Net or not) created on your system.