Eucalyptus versions 1.4 and higher includes a highly configurable VM networking subsystem that can be adapted to a variety of network environments. There are three high level networking "modes", each with its own set of configuration parameters, features, benefits and in some cases restrictions placed on your local network setup. The administrator must select one of these three modes before starting Eucalyptus on the front-end and nodes via modification of the 'eucalyptus.conf' configuration file on each machine running a Eucalyptus component. Brief descriptions of each mode follows:

SYSTEM Mode - This is the simplest networking mode, but also offers the smallest number of networking features. In this mode, Eucalyptus simply assigns a random MAC address to the VM instance before booting and attaches the VM instance's ethernet device to the physical ethernet through the node's local Xen bridge. VM instances typically obtain an IP address using DHCP, the same way any non-VM machine using DHCP would obtain an address. Note that in this mode, the Eucalyptus administrator (or the administrator that manages the network to which Eucalyptus components are attached) must set up a DHCP server that has a dynamic pool of IP addresses to hand out as VMs boot. In other words, if your laptop/desktop/server gets an IP address using DHCP on the same network as the Eucalyptus nodes, then your VMs should similarly obtain addresses. This mode is most useful for users who want to try out Eucalyptus on their laptops/desktops.

STATIC Mode - This mode offers the Eucalyptus administrator more control over VM IP address assignment. Here, the administrator configures Eucalyptus with a 'map' of MAC address/IP Address pairs. When a VM is instantiated, Eucalyptus sets up a static entry within a Eucalyptus controlled DHCP server, takes the next free MAC/IP pair, assigns it to a VM, and attaches the VMs ethernet device to the physical ethernet through the Xen bridge on the nodes (in a manner similar to SYSTEM mode). This mode is useful for administrators who have a pool of MAC/IP addresses that they wish to always assign to their VMs.

NOTE - Running Eucalyptus in SYSTEM or STATIC mode disables some key functionality such as the definition of ingress rules between collections of VMs (termed security groups in Amazon EC2), the user-controlled, dynamic assignment of IPs to instances at boot and run-time (elastic IPs in Amazon EC2), and isolation of network traffic between VMs (that is, the root user within VMs will be able to inspect and potentially interfere with network traffic from other VMs).

MANAGED Mode - This mode is the most featureful of the three modes, but also carries with it the most potential constraints on the setup of the Eucalyptus administrator's network. In MANAGED mode, the Eucalyptus administrator defines a large network (usually private, unroutable) from which VM instances will draw their IP addresses. As with SYSTEM mode, Eucalyptus will maintain a DHCP server with static mappings for each VM instance that is created. Eucalyptus users can define a number of 'named networks', or 'security groups', to which they can apply network ingress rules that apply to any VM that runs within that 'network'. When a user runs a VM instance, they specify the name of such a network that a VM is to be a member of, and Eucalyptus selects a subset of the entire range of IPs that other VMs in the same 'network' can reside. A user can specify ingress rules that apply to a given 'network', such as allowing ping (ICMP) or ssh (TCP, port 22) traffic to reach their VMs. This capability allows Eucalyptus expose a capability similar to Amazon's 'security groups'. In addition, the administrator can specify a pool of public IP addresses that users may allocate, then assign to VMs either at boot or dynamically at run-time. This capability is similar to Amazon's 'elastic IPs'. Eucalyptus administrators that require security groups, elastic IPs, and VM network isolation must use this mode.

Each Eucalyptus network mode has its own set of infrastructure requirements, configuration parameters, and caveats. These are described in more detail in the following sections.

SYSTEM Mode ¶

There is very little Eucalyptus configuration to use SYSTEM mode, as in this mode, Eucalyptus mostly stays 'out of the way' in terms of VM networking. The options in 'eucalyptus.conf' that must be configured correctly in 'SYSTEM' mode are as follows:

On the front-end:

VNET_MODE="SYSTEM"

On each node:

VNET_MODE="SYSTEM"
VNET_BRIDGE

In each Eucalyptus node controller's (NC) 'eucalyptus.conf' file, make sure that the parameter 'VNET_BRIDGE' is set to the name of the Xen bridge device that is connected to your local ethernet. In Xen 3.0 (and some other versions), the name of the bridge, by default, was 'xenbr0'. If you have such an installation, specify it like so:

VNET_BRIDGE="xenbr0"

In Xen 3.2 and higher, the name of the bridge (most of the time) is set to the name of your ethernet device (generally 'eth0'). If this is the case on your system, set configure Eucalyptus like so:

VNET_BRIDGE="eth0"

Make sure that what you are specifying in this field is actually a bridge, and that it is the bridge that is connected to an ethernet network that has a DHCP server running elsewhere that is configured to hand out IP addresses dynamically. Use the 'brctl show' command to inspect the status of your local bridges. Note that your front-end machine may not have any bridges if Xen is not installed (this is fine, as VNET_BRIDGE is only a relevant for node controllers, and will be safely ignored by the front-end components).

To test whether this mode is working properly at run-time, start an instance and log in to the node where the instance is running. Run 'xm list' to find the Xen ID that your instance is running under. Then, look at the output of 'brctl show', it should look something like this (assuming your VNET_BRIDGE is set to 'eth0', and the Xen ID of your instance was '18'):

; brctl show eth0
bridge name  bridge id          STP enabled     interfaces
eth0         8000.000c29369858  no              peth0
                                                vif18.0

note that Eucalyptus has correctly attached the VM's 'eth0' interface (vif18.0) to the bridge ('eth0') that is being used to attach VMs to the local ethernet ('peth0'). At this point, the VM should be sending DHCP requests to the local ethernet, and the DHCP server on the network should be sending a reply.

CAVEATS - In this mode, as mentioned previously, VMs are simply started with their ethernet interfaces attached to the local ethernet without any isolation. Practically, this means that you should treat a VM the same way that you would treat a non-VM machine running on the network. Eucalyptus does it's best to discover the IP address that was assigned to a running VM via a third-party DHCP server, but can be unsuccessful depending on the specifics of your network (switch types/configuration, location of CC on the network, etc.). Practically, if Eucalyptus cannot determine the VM's IP, then the user will see '0.0.0.0' in the output of 'describe-instances' in both the private and public address fields. The best workaround for this condition is to instrument your VMs to send some network traffic to your front end on boot (after they obtain an IP address). For instance, setting up your VM to ping the front-end a few times on boot should allow Eucalyptus to be able to discover the VMs IP.

STATIC Mode ¶

In this mode, Eucalyptus will manage VM IP address assignment by maintaining its own DHCP server with one static entry per VM. The options in 'eucalyptus.conf' that must be configured correctly in 'STATIC' mode are as follows:

On the front-end (options annotated with a '*' may be required depending on your installation, see below for details):

VNET_MODE="STATIC"
VNET_INTERFACE
VNET_DHCPDAEMON
*VNET_DHCPUSER
VNET_SUBNET
VNET_NETMASK
VNET_BROADCAST
VNET_ROUTER
VNET_DNS
VNET_MACMAP

On each node:

VNET_MODE="STATIC"
VNET_BRIDGE

The Eucalyptus administrator must configure the front-end's 'eucalyptus.conf' first with a valid, configured ethernet device that is attached to the same physical ethernet as the Eucalyptus nodes:

VNET_INTERFACE="eth0"

Next, the admin ust ensure that there is a DHCP server binary installed on the front-end and Eucalyptus knows where it is located:

VNET_DHCPDAEMON="/usr/sbin/dhcpd3"

If your DHCP daemon binary is configured to run as 'non-root' (say, as the user 'dhcpd' as is the case in Ubuntu >= 8.10), then you must configure Eucalyptus to be aware of that user:

VNET_DHCPUSER="<dhcpusername>"

Then, the admin must input IP subnet information for that device. For example, if the front-end's 'eth0' interface has the IP address '192.168.1.254' on the '192.168.1.0/24' network, with a gateway at '192.168.1.1' and a DNS at '192.168.1.2', the values in 'eucalyptus.conf' would look like so:

VNET_SUBNET="192.168.1.0"
VNET_NETMASK="255.255.255.0"
VNET_BROADCAST="192.168.1.255"
VNET_ROUTER="192.168.1.1"
VNET_DNS="192.168.1.2"

Finally, the administrator must supply a list of static MAC/IP mappings that will be assigned, first come first served, to VM instances. Note that each IP must reside in the subnet defined above, and must not be in use by any other machine on the network.

VNET_MACMAP="AA:DD:11:CE:FF:ED=192.168.1.3 AA:DD:CE:FF:EE=192.168.1.4"

On the nodes, you must ensure that the bridge is entered (typically 'xenbr0' on Xen 3.0 and 'eth0' on Xen >= 3.2). Run the command 'brctl show' to inspect your nodes' bridge setup.

VNET_BRIDGE="xenbr0"

Once you have configured Eucalyptus properly, start up the node controllers and the front-end components. To test whether this mode is working properly at run-time, start an instance and log in to the node where the instance is running. Run 'xm list' to find the Xen ID that your instance is running under. Then, look at the output of 'brctl show', it should look something like this (assuming your VNET_BRIDGE is set to 'eth0', and the Xen ID of your instance was '18'):

; brctl show eth0
bridge name  bridge id          STP enabled     interfaces
eth0         8000.000c29369858  no              peth0
                                                vif18.0

note that Eucalyptus has correctly attached the VM's 'eth0' interface (vif18.0) to the bridge ('eth0') that is being used to attach VMs to the local ethernet ('peth0'). Make sure that the DHCP server has been started properly on the front-end ('ps axww | grep -i dhcpd | grep -i euca'). At this point, the VM should be sending DHCP requests to the local ethernet, and the DHCP server on the front-end should be sending a reply with one of the static MAC/IP mappings the admin has defined in 'eucalyptus.conf'.

CAVEATS - In this mode, as mentioned previously, VMs are started with their ethernet interfaces attached to the local ethernet without any isolation. Practically, this means that you should treat a VM the same way that you would treat a non-VM machine running on the network. Eucalyptus does not verify that your settings are valid, thus, you must enter them correctly in order for your VMs to obtain IP addresses. Finally, we assume that the installed DHCP daemon is, or is compatible with, ISC DHCP Daemon version 3.0.X. If it is not, we recommend either installing a version that is (common in most distributions) or writing a wrapper script around your installed DHCP server and point Eucalyptus at it (via VNET_DHCPDAEMON in 'eucalyptus.conf').

MANAGED Mode ¶

In this mode, Eucalyptus will fully manage the local VM instance network and provides all of the networking features that Eucalyptus currently supports (VM network isolation, user controllable VM firewalls (ingress rules/security groups), dynamic public IP assignment). The options in 'eucalyptus.conf' that must be configured correctly in 'MANAGED' mode are as follows:

On the front-end (options annotated with a '*' may be required depending on your installation, see below for details):

VNET_MODE="MANAGED"
VNET_INTERFACE
VNET_DHCPDAEMON
*VNET_DHCPUSER
VNET_SUBNET
VNET_NETMASK
VNET_DNS
VNET_ADDRSPERNET
*VNET_PUBLICIPS

On each node:

VNET_MODE="MANAGED"
VNET_INTERFACE

Be advised that this mode requires that your local network/configuration conforms to certain requirements that Eucalyptus depends upon.

Requirements for MANAGED mode ¶

Before using 'MANAGED' mode, you must confirm that:

1.) there is an available range of iP addresses that is completely unused on the network (192.168..., 10....., other).

2.) your network is 'VLAN clean', meaning that all switch ports that Eucalyptus components are connected to will allow and forward VLAN tagged packets.

3.) you are not running a firewall on the front-end (CC) or your firewall is compatible with the dynamic changes that Eucalyptus will make to the front-end's netfilter rules.

All three of these requirements must be met before MANAGED mode should be attempted. Failure to verify the above will, at least, result VM instances being unavailable on the network.

For requirement '1', choose a IP range that you know is completely unused on your network. Choose a range that is as large as possible. Typical examples are:

if the network 10.0.0.0 - 10.255.255.255 is completely unused:

VNET_MODE="MANAGED"
VNET_SUBNET="10.0.0.0"
VNET_NETMASK="255.0.0.0"
VNET_DNS="<your DNS>"
VNET_ADDRSPERNET="128"

or if the network 192.168.0.0 - 192.168.255.255 is completely unused:

VNET_MODE="MANAGED"
VNET_SUBNET="192.168.0.0"
VNET_NETMASK="255.255.0.0"
VNET_DNS="<your DNS>"
VNET_ADDRSPERNET="64"

Next, the admin must verify that the local network will allow/forward VLAN tagged packets between machines running Eucalyptus components. To verify, perform the following test:

on the front-end, choose the interface that is on the local ethernet (and will be set in eucalyptus.conf as VNET_INTERFACE), and run:

vconfig add <interface> 10
ifconfig <interface>.10 192.168.1.1 up

replace '192.168.1.1' with an IP from the range you selected above.

On the node, choose the interface on the local network (will be set in eucalyptus.conf as VNET_INTERFACE), and run:

vconfig add <interface> 10
ifconfig <interface>.10 192.168.1.2 up

again, replace '192.168.1.2' with another IP in the range you selected above.

Then, try a ping between hosts. On the front-end:

ping 192.168.1.2

on the node:

ping 192.168.1.1

If this does not work, then your switch needs to be configured to forward VLAN tagged packets (if it is a managed switch, see your switch's documentation to determine how to do this).

Finally, you need to carefully inspect the firewall on the front-end to make sure that it will not interfere with Eucalyptus, or vice-versa. Eucalyptus will flush the 'filter' and 'nat' tables upon boot in MANAGED mode, but provides a way for the administrator to define special rules that are loaded when Eucalyptus starts (see below for details).

Configuring MANAGED mode ¶

The Eucalyptus administrator must configure the front-end's 'eucalyptus.conf' first with a valid, configured ethernet device that is attached to the same physical ethernet as the Eucalyptus nodes:

VNET_INTERFACE="eth0"

Next, the admin ust ensure that there is a DHCP server binary installed on the front-end and Eucalyptus knows where it is located:

VNET_DHCPDAEMON="/usr/sbin/dhcpd3"

If your DHCP daemon binary is configured to run as 'non-root' (say, as the user 'dhcpd' as is the case in Ubuntu >= 8.10), then you must configure Eucalyptus to be aware of that user:

VNET_DHCPUSER="<dhcpusername>"

Nodes must have VNET_INTERFACE set properly. For example, with current Xen versions, this parameter (when your node's Xen bridge is 'eth0') is typically:

VNET_INTERFACE="peth0"

Once you have verified that your network configuration meets the requirements for running in MANAGED mode, the rest of the configuration is fairly simple. For example, if the 192.168.0.0/16 network is free and unused on your network:

VNET_MODE="MANAGED"
VNET_SUBNET="192.168.0.0"
VNET_NETMASK="255.255.0.0"
VNET_DNS="<your dns>"
VNET_ADDRSPERNET="64"
VNET_PUBLICIPS="<publicIPa> <publicIPb> ... <publicIPn>"

SUBNET, NETMASK, and DNS have been described previously. VNET_ADDRSPERNET is used to control how many VM instances may simultaneously be part of an individual user's named network (called a 'security group' in Amazon EC2). Choosing the right value for this parameter depends on how many IPs you have made available using VNET_SUBNET/VNET_NETMASK, how many VLANs your network supports simultaneously, and how many concurrent active user networks the administrator wishes to support. In the above example, there are 65536 addresses available (192.168.0.0/16). If we divide by the number of addresses per network (set to 64 above), we find the maximum number of simultaneous active named networks that can be in use at any one point in time (65536 / 64 == 1024). If your eucalyptus installation has 100 users, then each user could have at most 10 active security groups in operation at any point in time (of course, they can define as many as they wish, but can only have sets of running VMs residing in at most 10 networks). Each security group could support up to 61 instances (64 addresses minus 1 address for the subnet, broadcast, and router IPs). If your installation favors more VMs per network and fewer active security groups per user, the administrator may adjust the VNET_ADDRSPERNET parameter accordingly. Setting it to '256' would result in each active user's security group supporting up to 253 VM instances, and each of 100 users could simultaneously have 2 active security groups.

If you would like users to log in to their instances from outside the cluster/cluster front-end, you must find a set of public IP addresses, that are not in use, and allow Eucalyptus to dynamically route them to VM instances at instance boot time or dynamically at run time. For each IP address you choose, your front-end must be capable of being configured with that IP address. To test, choose some free public IP addresses and perform the following test for each one:

on the front-end:

ip addr add <publicIP>/32 dev <interface>

on some external machine representative of where users will wish to log into their VM instances:

ping <publicIP>

if this works, then dynamic IP assignment to VM instances will work. Remove the assigned address with the following command:

ip addr del <publicIP>/32 dev <interface>

Once you have compiled a list of available public IP addresses, allow Eucalyptus to use them by listing the IPs in 'eucalyptus.conf':

VNET_PUBLICIPS="<publicIPa> <publicIPb> ... <publicIPn>"

CAVEATS - When Eucalyptus is running in MANAGED mode, you cannot currently run an entire eucalyptus installation o n a single machine as this mode depends upon traffic between named networks passing through a front-end router (instead of going through the loopback device). If you wish to run Eucalyptus on a single machine (laptop), you must use SYSTEM or STATIC mode. In MANAGED mode, Eucalyptus will flush the front-end's iptables rules for both 'filter' and 'nat'. Next, it will set the default policy for the 'FORWARD' chain in 'filter' to 'DROP'. At run time, the front-end will be adding and removing rules from 'FORWARD' as users add/remove ingress rules from their active security groups. In addition, the 'nat' table will be configured to allow VMs access to the external network using IP masquerading, and will dynamically add/remove rules in the 'nat' table as users assign/unassign public IPs to VMs at instance boot or run-time. If the administrator has some rules that they wish to apply o the front-end, they should perform the following procedure on the front-end, before eucalyptus is started or while eucalyptus is not running. WARNING if the admin chooses to perform this operation to define special iptables rules that are loaded when Eucalyptus starts, they could inadvertently cause Eucalyptus VM networking to fail. It is suggested that you only do this only if you are completely sure that it will not interfere with the operation of Eucalyptus.

<use iptables to set up your iptables rules>
iptables-save > $EUCALYPTUS/var/run/eucalyptus/net/iptables-preload

Troubleshooting MANAGED Mode ¶

If you start an instance believe that it is running but is not available on the network, here are some things to check.

First, verify that the requirements of MANAGED mode have been met as described above (unused range of IPs, VLAN capable network, no interfering firewall rules on the nodes or front-end). Test whether you can get to the instance from the front-end using it's private address (from the range you specified). If you cannot, next, inspect the interfaces on the front-end and nodes:

on front-end:

ifconfig -a

You should see an interface '<interface>.<vlan>' with an IP address that is up and running. For instance, if may be 'eth0.10'. If it is not, check your VNET_INTERFACE parameter and inspect the eucalyptus log files for errors.

on the node:

brctl show

You should see a number of bridges called 'eucabr<vlan>', where '<vlan>' is a number that typically starts from '10'. The output should be similar (if VNET_INTERFACE="peth0") to:

; brctl show eucabr10
bridge name  bridge id          STP enabled     interfaces
eucabr10     8000.000c29369858  no              peth0.10
                                                vif18.0

If this is not the case, check your VNET_INTERFACE setting, and inspect the logfiles for details.

Back on the front-end, make sure that 'dhcpd' is running:

ps axww | grep <dhcpd>

where '<dhcpd>' is what you have set for VNET_DHCPDAEMON. Make sure that, in the output of 'ps', you see that the daemon is listening on the vlan tagged interface from above (<interface>.<vlan>). If it is not running, check the eucalyptus logs for the reason why (if the command failed, you will see this information in 'cc.log', if the daemon failed at runtime, you can inspect the reason in the daemon's output itself in 'http-cc_error_log'.

If you can access the private IP of the instance from the front-end, but public IPs are not being forwarded properly, first confirm that the user's security group is set up properly by having them run 'ec2-describe-group <group of instance>'. '<group of instance>' is set to 'default' by default or if unspecified when the instance was started. If the group has appropriate ingress rules set, check that the rules have been implemented on the front-end:

iptables -L <username>-<groupname>

If there are no rules here, check the 'cc.log' for errors applying the table rules for more insight. Next, check the 'nat' table:

iptables -L -t nat

You should see one DNAT rule for routing traffic from a public IP to the instance IP, and one SNAT rule for setting the source IP of outgoing packets from that instance. If you do not, check 'cc.log' to determine the cause.

If all of these checks pass and the instance still is experiencing network problems, please prepare the following information and send it along to the Eucalyptus discussion board:

on front-end and one representative node, capture the output of the following commands:

netstat -rn
ifconfig -a
brctl show
iptables-save

and send us 'cc.log', 'nc.log', 'httpd-cc_error_log' and 'httpd-nc_error_log'.