[Xen-users] What makes live migration so slow?

[Wenda Ni <wonda.ni@xxxxxxxxx>]

Hello,

I have now shfited to Centos 5.5, and I am testing live migration between 2 physical hosts with XEN 3.1.2. XEN 3.1.2 (virtualization) is included in centOS 5.5 during the installation phase, so everything is handled by default. A third host with Ubuntu OS serves as the network file system. The three hosts are connected by one D-link gigabit switch (DGS-2205).

The downtime of live migration is still very long. If we measure the downtime by Ping command,  one migration event of a guest VM (domU) gives a loss of 30-50 packets. (should be over half a minute).

Could anyone know the potential problem? Thank you in advance!


The xend-config.sxp is attached as follows, and the firewall has been disabled on port 8000, 8002 (both tcp & udp).

# -*- sh -*-

#
# Xend configuration file.
#

# This example configuration is appropriate for an installation that 
# utilizes a bridged network configuration. Access to xend via http
# is disabled.  

# Commented out entries show the default for that entry, unless otherwise
# specified.

#(logfile /var/log/xen/xend.log)
#(loglevel DEBUG)

#(xend-http-server no)
(xend-http-server yes)
(xend-unix-server yes)
#(xend-tcp-xmlrpc-server no)
#(xend-unix-xmlrpc-server yes)
(xend-relocation-server yes)
# The relocation server should be kept desactivated unless using a trusted
# network, the domain virtual memory will be exchanged in raw form without
# encryption of the communication. See also xend-relocation-hosts-allow option

(xend-unix-path /var/lib/xend/xend-socket)

# Port xend should use for the HTTP interface, if xend-http-server is set.
(xend-port            8000)

# Port xend should use for the relocation interface, if xend-relocation-server
# is set.
(xend-relocation-port 8002)

# Address xend should listen on for HTTP connections, if xend-http-server is
# set.
# Specifying 'localhost' prevents remote connections.
# Specifying the empty string '' (the default) allows all connections.
(xend-address '')
#(xend-address localhost)

# Address xend should listen on for relocation-socket connections, if
# xend-relocation-server is set.
# Meaning and default as for xend-address above.
(xend-relocation-address '')

# The hosts allowed to talk to the relocation port.  If this is empty (the
# default), then all connections are allowed (assuming that the connection
# arrives on a port and interface on which we are listening; see
# xend-relocation-port and xend-relocation-address above).  Otherwise, this
# should be a space-separated sequence of regular expressions.  Any host with
# a fully-qualified domain name or an IP address that matches one of these
# regular expressions will be accepted.
#
# For example:
#  (xend-relocation-hosts-allow '^localhost$ ^.*\.example\.org$')
#
(xend-relocation-hosts-allow '')
#(xend-relocation-hosts-allow '^localhost$ ^localhost\\.localdomain$')

# The limit (in kilobytes) on the size of the console buffer
#(console-limit 1024)

##
# To bridge network traffic, like this:
#
# dom0: fake eth0 -> vif0.0 -+
#                            |
#                          bridge -> real eth0 -> the network
#                            |
# domU: fake eth0 -> vifN.0 -+
#
# use
#
# (network-script network-bridge)
#
# Your default ethernet device is used as the outgoing interface, by default. 
# To use a different one (e.g. eth1) use
#
# (network-script 'network-bridge netdev=eth1')
#
# The bridge is named xenbr0, by default.  To rename the bridge, use
#
# (network-script 'network-bridge bridge=<name>')
#
# It is possible to use the network-bridge script in more complicated
# scenarios, such as having two outgoing interfaces, with two bridges, and
# two fake interfaces per guest domain.  To do things like this, write
# yourself a wrapper script, and call network-bridge from it, as appropriate.
#
(network-script network-bridge)

# The script used to control virtual interfaces.  This can be overridden on a
# per-vif basis when creating a domain or a configuring a new vif.  The
# vif-bridge script is designed for use with the network-bridge script, or
# similar configurations.
#
# If you have overridden the bridge name using
# (network-script 'network-bridge bridge=<name>') then you may wish to do the
# same here.  The bridge name can also be set when creating a domain or
# configuring a new vif, but a value specified here would act as a default.
#
# If you are using only one bridge, the vif-bridge script will discover that,
# so there is no need to specify it explicitly.
#
(vif-script vif-bridge)


## Use the following if network traffic is routed, as an alternative to the
# settings for bridged networking given above.
#(network-script network-route)
#(vif-script     vif-route)


## Use the following if network traffic is routed with NAT, as an alternative
# to the settings for bridged networking given above.
#(network-script network-nat)
#(vif-script     vif-nat)


# Dom0 will balloon out when needed to free memory for domU.
# dom0-min-mem is the lowest memory level (in MB) dom0 will get down to.
# If dom0-min-mem=0, dom0 will never balloon out.
(dom0-min-mem 256)

# In SMP system, dom0 will use dom0-cpus # of CPUS
# If dom0-cpus = 0, dom0 will take all cpus available
(dom0-cpus 0)

# Whether to enable core-dumps when domains crash.
#(enable-dump no)

# The tool used for initiating virtual TPM migration
#(external-migration-tool '')

# The interface for VNC servers to listen on. Defaults
# to 127.0.0.1  To restore old 'listen everywhere' behaviour
# set this to 0.0.0.0
#(vnc-listen '127.0.0.1')
  
# The default password for VNC console on HVM domain.
# Empty string is no authentication.
(vncpasswd '')

# The default keymap to use for the VM's virtual keyboard
# when not specified in VM's configuration
(keymap 'en-us')

# The VNC server can be told to negotiate a TLS session
# to encryption all traffic, and provide x509 cert to 
# clients enalbing them to verify server identity. The
# GTK-VNC widget, virt-viewer, virt-manager and VeNCrypt
# all support the VNC extension for TLS used in QEMU. The
# TightVNC/RealVNC/UltraVNC clients do not.
#
# To enable this create x509 certificates / keys in the
# directory /etc/xen/vnc
#
#  ca-cert.pem       - The CA certificate
#  server-cert.pem   - The Server certificate signed by the CA
#  server-key.pem    - The server private key
#
# and then uncomment this next line
# (vnc-tls 1)
#
# The certificate dir can be pointed elsewhere..
#
# (vnc-x509-cert-dir /etc/xen/vnc)
# 
# The server can be told to request & validate an x509
# certificate from the client. Only clients with a cert
# signed by the trusted CA will be able to connect. This
# is more secure the password auth alone. Passwd auth can
# used at the same time if desired. To enable client cert
# checking uncomment this:
#
# (vnc-x509-verify 1)

# Allow probing of disk image file format.  This is insecure!  It lets
# a malicious domU read any file in dom0.  Applies only to fully
# virtual domUs.  Required for using formats other than raw.
#(enable-image-format-probing no)

# Number of seconds xend will wait for device creation
#(device-create-timeout 100)

# Strict checking when doing PCI passthrough; enabled by default
#(pci-dev-assign-strict-check yes)


--

Cheers,

 

Wenda Ni, Ph.D.

Dept. of Computer Science and Engineering

State University of New York (SUNY) at Buffalo

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users