Deployment of an image kills the network...
We're managing roughly thirty computers and regularly need to reimage the machines, as it's a lab enviornment. Nevertheless, upon the deployment of an image to one lab (roughly 13-16 PC's), the network becomes completely unusable for both lab enviornments. Therefore, we must wait until the end of the day to reimage.
Right now the network is situated so that both lab enviornments share a single T1, which is plugged into a Netgear FVS318 Router; the router handles all DCHP for the lab. The Ghost Server is directly plugged into the Netgear Router and there are two Linksys SR224 swiches, which handle both lab enviornments.
Should the Ghost server be plugged into one of the two classroom switches, rather than directly into the router? Is there another configuation of the network that would work better? Is there a setting which I need to modify within the router like RIP, or RIP Directon?
Thanks in advance,
CF
Comments
With that gear, use unicast or cap the transmit rate
That networking equipment is pretty low-end - although the switches have a lot of ports, they simply aren't capable of handling much load, and there's nothing that can really be done about than either not using Ghost's default multicast mode of operation and specifying Unicast transmission (which will be quite slow with only 100mps, but won't impact the other switch ports), or dialling back Ghost's transmit rate to the point where the switches can cope. If they were unmanaged gigabit switches they'd probably cope better but using 100mbps gear for this is pushing it.
A network with unmanaged switches means the switches won't be aware of IP multicast and will just try and stuff all the Ghost traffic down every port, and very low-end switches may not handle high offered loads well in this case. Standard TCP/IP multicast in switched Ethernet works best when you use switches that support it (using a feature called IGMP snooping, in concert with a router that itself supports IGMP and can produce IGMP query packets), in which case only the switch ports participating in the Ghost session receive the traffic, while the Ghost server still only sends each frame once and so the link from the server to the switch doesn't get saturated as it does with Unicast.
To get an idea of the switches that work best with Ghost at that kind of port count, have a look at this HP Procurve selection guide (which helpfully lists the series numbers for other vendors in the same categories, so you can compare them up; look for switches with IGMP support, which means a managed switch like the Procurve 2510G, Linksys SRW 2024, 3Com 4200G or Cisco Catalyst Express is the level of gear to aim for (in conjunction with a suitable router to generate the IGMP queries for the switches to snoop).
Would you like to reply?
Login or Register to post your comment.