Ghosting Machines across subnets
Hello I am planning to use ghost to upgrade some computers with new images in the near future. Currently our ghost server is in one building with a DHCP server and the computers I need to upgrade are in a separate building with its own DHCP server (both are with in a block of each other). And I would like to upgrade more than one computer at a time if possible but I need to keep network traffic to a minimum as there are some systems that can not be interrupted or have network bogged down. I have been doing some interrupted research on this problem for a bit and would just like some clarification.
My first question is with multicast and how much that will affect the network with traffic? I am also curious as how direct cast works and whether it causes a lot of network traffic on the targeted subnet? Lastly is there any way to use unicast in this situation?
I am kind of expecting to have to do many unicasts since primary concern is that the devices on the network need to function properly with out interruption.
Thank you in advance for any help anyone is able to provide.
Comments
Sorry for not posting an answer earlier, but I've been unable to post for the past 5 days due to the forum servers being misconfigured.
This is not something that is possible to say in the abstract; the results you get are entirely dependent on the network equipment you have installed, and the ability of your operations staff to configure and run it.
In principle, multicast is always the most efficient mode of operation for network traffic like Ghosts's (which is indeed why, with the support of the equipment makers, many core protocols in IPv6 use multicast). Broadcast bus designs like classical DIX Ethernet (switched Ethernet being somewhat different) naturally support multicast efficiently, and were widely employed 20 years ago when IP multicasting was defined as an IETF standard.
However, during the migration from classical broadcast Ethernet to switched Ethernet, the capabilities of switch equipment took a very long time to reach a sufficient level of capability to be able to achieve the same capabilities with respect to multicast as the original designs had. Most unmanaged switches simply treat multicast frames as broadcast frames that are sent to all switch ports, for instance.
Most managed switches of the past decade are capable of doing rather better, and indeed nowadays even quite cheap switches suitable for small businesses are capable of using the same techniques, where the switches listen to the standard IGMP traffic between endpoints and routers and from that learn which switch ports subscribe to which multicast frames.
Now, in your case what you need to understand is what equipment (what models, from what manufacturer) is present at every step between the buildings to know whether it has been configured with this support. It should be, but since IP multicast even 20 years on is not something many people plan for, it is possible that your network is not set up for it at all. You do need to discuss this with your network operations people.
Basically, all Ghost needs to work is a network that provides standard IP multicast connectivity (that's our equivalent of "dial tone" ) via IGMP subscriptions. The choice of configuration options deeper in the network - advanced things like alternate PIM modes - aren't things we specify because Ghost should in principle work with any of them - it's really up to the precise capabilities of the network equipment and the operational needs of the wider network that should determine things like that.
It works as you would expect; a single frame will be sent from the server to target the destination subnet. When the frame enters that subnet, it will be sent broadcast to all the machines in that subnet (and most switches will attempt to send that frame out every switch port).
Whether this is "a lot of traffic" isn't again something that can be stated in the abstract, however. Most modern networks, gigabit-equipped and with normal commercial-quality switches and hosts, will have few issues with this. Networks with consumer-grade switches or old consumer-grade hosts (and especially with older 10Mbit or half-duplex connections on them) may have extreme difficulty and indeed the network equipment may fail.
[ There is one minor caveat; many systems nowadays have low-power idle modes, and it is not uncommon in such environments for the network connection between host and switch to negotiate a 10Mbps rate. There may be some problems using directed broadcast in such environments since the switches will be unable to broadcast frames out ports which have negotiated such a slow rate. ]
In a wider cross-subnet situation, unicast is problematic. Instead of the server sending one frame and having copied of that made efficiently by the network leading to traffic loads proportional to the size of the image, instead in unicast mode each endpoint receives an individual copy of everything originated from the server. So, for 30 machines the traffic load on the intervening network is 30 times greater than for multicast or even directed broadcast would be.
It is always available as a last resort. However, it is only something to be used as a last resort, because it puts such immense stress on the network infrastructure. It's only really suitable within a single subnet, where due to configuration problems in the wider network multicast isn't available (but the switching capabilities of even low-end switches are usually adequate to cover the deficiencies of using unicast).
Ghosting across subnets
Ok Nigel is it possible to ghost across subnets, the issue Im running into is the initial connection. My client machines are not connecting to the server to get the VBF and from there nothing else can happen. Any ideas???
It's always been possible to
It's always been possible to use Ghost across subnets; since GhostCast uses completely standard IPv4 multicast traffic, all that is necessary is to have correctly configured multicast routing between the subnets, a topic that pretty much any router (or managed switch) vendor should cover in their product documentation (multicast routers will also generally function as IGMP queriers for their directly attached networks). If your network administrator is unfamiliar with IPv4 multicast concepts then the key protocol family they need to learn about is called PIM, for Protocol Independent Multicast.
Would you like to reply?
Login or Register to post your comment.