I have DPDK set up, with two NICs taken over by DPDK, i.e. the interfaces
are not visible to userspace applications.
Now, I need to run netperf/iperf to measure throughput performance of port-to-port
configuration of DPDK, as described here https://github.com/01org/dpdk-ovs/blob/development/docs/04_Sample_Configurations/00_Phy-Phy.md
However netperf server requires to be bound to the network interface, and netperf client
has to know the netserver's IP address. But since the interfaces were taken
by DPDK, and netperf can't see their IP addresses any longer, how could I
run tests? Is there a way to redirect netperf right in network port,
regardless of IP addressess.
iptables, ebtables.. ?
Thanks.
If you're going by that diagram, your netperf and netserver should both be on the traffic generator. On the DPDK host, traffic coming in one interface is simply routed out the other interface by OVS, and back to your generator. There are a variety of tricks to get your generator to send traffic out on the wire rather than short circuiting over loopback, but they have been addressed here already.
If what you really want to do is have netserver running on the DPDK host, then you simply need to create a virtual interface, attach it to OVS, give it an IP, and direct your netperf traffic to it.
Related
I have one or more virtual machines on Debian host and two physical eth interfaces. I want to split bandwidth between eths (both for downlink and one for uplink). Is it possible with openvswitch and openflow?
The short answer is that it should be possible with OVS and OpenFlow. With OVS you can connect your VM's virtual ports and the server's physical interfaces.
Without thinking too much, you can load balancing the traffic by:
Installing a flow to direct any VM packet to your uplink port. This flow should rewrite the src IP and MAC as the ones from the downlink interface, so that it will look like it is being sent through that port.
Keep in mind that you might take your virtual ports configuration into account, and that you need some kind of mapping (something like NAT), to get the packets correctly returned to its respective VM. You can take a look in a NAT implementation, for the Ryu controller, to get some inspiration.
Running on a Linux system, getting UDP packets from another computer address to let's say 192.168.0.2 from another address let's say 192.168.166.66, I can see the UDP packets coming in with tcpdump. However, if I use netcat I don't actually receive the packets.
If I create an interface on 192.168.166.XXX network, then netcat is able to receive the packets no problem.
What basic networking concept am I missing? Why do I need to have an interface on the network of the sending IP when I can see with tcpdump that they are being delivered correctly?
tcpdump per default puts the interface into promiscious mode, which lets you see all the packets arriving at your network interface. But, your operating system only processes packets destined for the local system, e.g. either having the local or a broadcast address as destination.
The final solution to this problem was to disable Reverse Path Forwarding (RPF) on the interface. There are security implications here, but after careful review this was the correct path forward in this particular case.
RPF was turned off by modifying /etc/sysctl.conf:
net.ipv4.conf.eth0.rp_filter=0
Some more information on RPF:
Wikipedia - Reverse path forwarding
Linux kernel rp_filter settings
I've been investigating networking for use in a two-player game I'm writing, and I'm still not clear on when a device must have a port forwarded in order to communicate with the outside world.
From what I've seen in other games, port forwarding is always required in order to host a server, but is not required on the client. In addition, there are other situations, such as skype (which, to my understanding is ultimately client to client), where neither end must forward a port.
So my question is, in over-the-Internet communication, when is and isn't port forwarding necessary, and what steps can i take as a developer to make it so my users don't have to worry about it? Thanks in advance!
Port forwarding is needed when a machine on the Internet needs to initiate a connection to a machine that's behind a firewall or NAT router. If the connection is initiated by the machine behind the firewall, the firewall/router automatically recognizes the reply traffic and sends it to the machine that opened the connection.
But if a packet arrives on the external interface, and it's not a part of such a connection, the router needs to know what to do with it. By default, it will reject it. But if forwarding is configured for the port, that tells it what internal machine to send it to.
Put another way: you need port forwarding if you want to run a server behind the NAT firewall/router, you don't need it if you're just running a client.
There is reason why Skype don't (not always) need manual setting of port forwarding:
When you install Skype, a port above 1024 is chosen at random as the
port for incoming connections. You can configure Skype to use a
different port for incoming connections if you wish, but if you do,
you must open the alternative port manually.
If the port chosen for incoming connections becomes unavailable, by
default ports 80 and 443 will be used as alternatives. If another
application (such as Apache HTTP server or IIS) uses these ports, you
can either configure the application to use other ports, or you can
configure Skype to not use these ports.
Port forwarding is must if you host a server.
You can use same technique as Skype...
I am not sure if there is any other option...
Port forwarding (occurs) when a NAT, firewall or some other device blocks communication on all or some ports.
To answer your question as an example, most commercial routers use NAT to allow multiple people to use the same IP(As view from the outside world) provided by ISPs. Most ISP's use NAT to allow multiple customers to use the same IP(As viewed from the outside world). To get this to work, the NAT changes the internal IP and the port number of a communication to THE(there is only one for the entire sub network) external IP and a new port number. By doing this, the router/isp/ect can tell which internal IP and port each external communication goes to.
Anytime one of the computers communicating over the internet are behind a NAT, port forwarding is required. I'm sure there are way more situations than this, and the solution to each can be quite complicated. But this covers the vast majority.
I am developing a networked application that runs on a few different computers on a LAN. One of the core needs is for the app to maintain a list of peers on the LAN with which it has communicated in the past, so that it can restore previous sessions. The naive solution would be to just remember the IP and store it in a table, but what happens when the IP of a peer changes?
Instead, I thought I'd store the hostname of the peers so even if the IP changes they will still be reachable via their hostname. (I know hostnames can change as well but that is good enough).
So my question is what exactly is needed to make hostname resolution work on a LAN with mixed Windows/Mac/Linux clients?
Without the use of a central authority the only reliable way to achieve this is through the use of zerconfiguration name resolution. This means that without a multicast router you will only be able to dynamically resolve peers on the same subnet as the resolving host. You could use something like bonjour for mac, netbios or ssdp for windows or avahi for linux but you can't assume that these are enabled. I may be overlooking some more popular protocols that perform this function well but I would personally throw together a quick udp broadcast name resolution protocol for your application. Take a look at these for some more ideas:
Zeroconf Name resolution
Universal local network name resolution method without DNS?
http://en.wikipedia.org/wiki/Zero_configuration_networking#Name_resolution
http://en.wikipedia.org/wiki/Broadcast_address#IP_networking
I would pick a specific udp port to listen on (lets say 12000) and then when you're ready to resolve hosts send a "hello" udp packet out to 255.255.255.255 on port 12000 and all of the other hosts on your network running your app should reply with a packet containing their hostname, possibly other information.
we have little network devices which are shipped with IP address 0.0.0.1 to ensure that they never collide with any other device in their new environment (thus none of the 10.x.x.x, 172.16.x.x or 192.168.x.x ranges) until configuration. DHCP is no solution since there might be no DHCP server in the field.
The devices would listen to UDP broadcasts and answer with broadcasts until they are given their new IP address this way.
This worked fine with Windows XP - but sucks with Windows 7: the config program does not receive the answer packets from the devices which still have 0.0.0.1. Wireshark sees the packets, then they are dumped by the system.
Question: Is there any reason (RFC?) that actually prohibits using this address in a local environment? Or is it just MS that was overcautious? Where can I read why they treat this address "invalid"? Which ranges are really "invalid" now, too?
Any idea of a workaround on the PC side (Win 7)?
I know that it is not recommended to use 0.xxx addresses for work places, but for this very reason - having a not-used address - it works perfectly.
Edit: there is a device out there called "Netburner" which might have faced the similar issue, according to their forum. See: http://forum.embeddedethernet.com/viewtopic.php?f=5&t=612&p=2198 Does - by coincidence - anybody know some background information?
It sounds as if your configuration application is listening for broadcast packets on all network interfaces and expecting to receive packets from foreign subnets.
That should not work - the OS should only pass-on broadcast packets from the subnets each network interface is on, not from all subnets on the same physical (e.g. Ethernet) segment. I am reasonably certain that doing otherwise is broken behaviour WRT the IP protocol.
The are two ways to deal with this:
Make sure that your network interface has an IP address in the target subnet. You can have more than one IP addresses for each network card, so that should not interfere with normal network operations.
Configure or modify you application to use raw sockets, like Wireshark. Keep in mind, however, that this overrides all normal checks and balances and should be avoided, since it can cause behaviour that is almost impossible to diagnose - which is why it is frowned upon by meny network administrators.
Can you you add new routing table entries to Windows machines easily? Windows has to know which interface to use when routing a broadcast packet to the 0.0.0.x network.
The Unix machines I'm familiar with have a routing table that maps network/netmask entries to either gateways or interfaces (if the network is a local network). The local network (192.168.0.0/16 for my home network) gets sent to interface eth0. Everything else 0.0.0.0/0 gets sent to a specific gateway machine 192.168.0.1.
If my machine sent a UDP broadcast message to network 0.0.0.0/24 (in other words, UDP broadcast sent to 0.0.0.255, then my machine would forward the packet to the gateway machine (which it can look up via arp). The switches in the middle wouldn't propagate the packet to other network devices, because the MAC address is set.
If my machine had another routing entry for 0.0.0.0/24 to the local interface, then my machine would send the packet on the wire using an ethernet broadcast group, and the switches would forward the packet to all connections. (Yay! Just like hubs in the 90s! :)
So I figure you need to add a routing entry for 0.0.0.0/24 to your client machines, so that they can properly address the broadcast packet.