User guide

11
Internet Group Management Protocol
The Internet Group Management Protocol
(IGMP) is designed to prevent multicast
flooding by allowing Layer 3 switches to
check whether host computers within their
care are interested in receiving particular
multicast transmissions. They can then
direct multicast data only to those points
that require it and can shut off a multicast
stream if the subnet has no recipients.
There are currently three IGMP versions:
1, 2 and 3, with each version building upon
the capabilities of the previous one:
• IGMPv1 allows host computers to opt
into a multicast transmission using a Join
Group message, it is then incumbent on
the router to discover when they no
longer wish to receive; this is achieved by
polling them (see IGMP Querier below)
until they no longer respond.
• IGMPv2 includes the means for hosts to
opt out as well as in, using a Leave Group
message.
• IGMPv3 encompasses the abilities of
versions 1 and 2 but also adds the ability
for hosts to specify particular sources of
multicast data.
AdderLinkInnityunitsmakeuseof
IGMPv2 when performing multicasts to
ensure that no unnecessary congestion is
caused.
IGMP Snooping
The IGMP messages are effective but only
operate at layer 3 - intended for routers
to determine whether multicast data
should enter a subnet. A relatively recent
development has taken place within the
switches that glue together all of the hosts
withineachsubnet:IGMPSnooping.IGMP
snooping means these layer 2 devices now
have the ability to take a peek at the IGMP
messages. As a result, the switches can
then determine exactly which of their own
hosts have requested to receive a multicast
– and only pass on multicast data to those
hosts.
IGMP Querier
When IGMP is used, each subnet requires
one Layer 3 switch to act as a Querier. In
this lead role, the switch periodically sends
out IGMP Query messages and in response
all hosts report which multicast streams
they wish to receive. The Querier device
and all snooping Layer 2 switches, then
update their lists accordingly (the lists are
also updated when Join Group and Leave
Group (IGMPv2) messages are received).
IGMP Fast-Leave (aka Immediate Leave)
Whenadevice/hostnolongerwishesto
receive a multicast transmission, it can
issue an IGMP Leave Group message as
mentioned above. This causes the switch
to issue an IGMP Group-SpecicQuery
message on the port (that the Leave
Group was received on) to check no other
receivers exist on that connection that
wish to remain a part of the multicast.
This process has a cost in terms of switch
processor activity and time.
Where ALIF units are connected directly
to the switch (with no other devices on the
same port) then enabling IGMP Fast-Leave
mode means that switches can immediately
remove receivers without going through
a full checking procedure. Where multiple
units are regularly joining and leaving
multicasts, this can speed up performance
considerably.
Jumbo frames (Jumbo packets)
Sinceitscommercialintroductionin1980,
theEthernetstandardhasbeensuccessfully
extended and adapted to keep pace with
the ever improving capabilities of computer
systems. The achievable data rates, for
instance, have risen in ten-fold leaps
fromtheoriginal10Mbit/stoacurrent
maximumof100Gbit/s.
While data speeds have increased
massively,thestandarddeningthenumber
of bytes (known as the Payload) placed into
each data packet has remained resolutely
stuck at its original level of 1500 bytes.
This standard was set during the original
speedera(10Mbits/s)andofferedthebest
compromise at that speed between the
time taken to process each packet and the
time required to resend faulty packets due
to transmission errors.
But now networks are much faster
andles/datastreamsaremuchlarger;
so time for a change? Unfortunately, a
wholesale change to the packet size is
not straightforward as it is a fundamental
standard and changing it would mean a
loss of backward compatibility with older
systems.
Larger payload options have been around
for a while, however, they have often
beenvendorspecicandatpresentthey
remainoutsidetheofcialstandard.There
is, however, increased consensus on an
optional‘Jumbo’payloadsizeof9000bytes
and this is fully supported by the AdderLink
Innity(ALIF)units.
Jumbo frames (or Jumbo packets)
offer advantages for ALIF units when
transmitting certain high resolution video
signals across a network. This is because
the increased data in each packet reduces
the number of packets that need to be
transferred and dealt with - thus reducing
latency times.
The main problem is that for jumbo frames
to be possible on a network, all of the
devices on the network must support
them.
Glossary
Glossary