Multicast speed issues
-
Problem: for a 100mbps roon, Unicast speeds show as about 780Mib/min, yet on multicast they drop to under 100Mib/min
A little snippet of the log, for imaging 30 machines
[QUOTE]
Timeout notAnswered=[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29] notReady=[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29] nrAns=0 nrRead=0 nrPart=30 avg=99
bytes= 8 164 796 640 re-xmits=0000133 ( 0.0%) slice=0058 73 709 551 615 [/QUOTE]
Seems to be a common issue, some people live with it or switch to unicast only but for the computers here, we would really like the use multicast, so I want to get to the heart of the issue.
Things to note:- The FOG server is on a 1gbps link. The current switch that many of the machines here are connected to are 100mbps links.
- Physical - and remote - access to the switches is somewhat limited, because of the way the organisation works here. Yet we have ~300 - 400 PCs to manage.
My thinking is, is that because the switches are only 100mpbs out, the buffer for the 100mbps could be overflowing with the 1gbps packets coming in. UDP just gets thrown out there and the poor 100mbps clients drop most of the stuff being thrown at them. As a result, FOG has to retransmit loads of data and thus takes more time with more clients, I would imagine (whereas, I guess, with more unicast clients it would take proportionally less time because of the additional load from multiple machines slowing down the rate to each client, meaning less dropped packets).
I could be wrong there, but why im posting under FOG issues is that I was wondering if there was a way to lets say “cap” how many packets are sent out at a time, or adjust the rate in fog (IE slow the speed down by adjusting the transmission rate of UDP packets from within FOG). Another solution would be to adjust the NIC rate I guess, but anything that would need to be done to the switches could be problematic, as it means talking to the people who manage our network equipment.
Any thoughts or comments about the above are most appreciated. And any help, of course!
-
Hey Trevelyan
Multicasting works as fast as the slowest link in the chain.
Imagine my disappointment when we just finished rolling out a brand new network infrastructure using HP 5406zl Switches with 10gb fibre connections between locations, only to find that a multicast of 30 computers resulted in a 500MiB/min speed in a room directly connected to the core switch!!!.I tracked it down to one computer on a dodgy patch in the wall and presto!!! 4.01GiB / Min
So a few things to check.
-
Does your switch support IGMP Snooping (if not, then the slow down can be from flooding the VLAN with traffic to every nic connected to it and awaiting the packet drop)
-
Do you suspect any of the PC’s you are imaging to has a dodgy connection or NIC?
there is no reason why multicast should not work as fast as unicast all things being equal.
-
-
Check to see what kind of other devices are plugged into the switches. I.E. Printers and turn them off, and try to multicast again. I have HORRIBLE speeds if I leave some of the MFD devices turned on while casting.
-
Lethal - Multicasting should be as fast as Unicast yes, but if FOG is sending packets via UDP at 1gpbs, the switch would surely drop a lot of the incoming packets because it can’t send them out of the 100mbps fast enough (and thus lots of retransmissions)? Even with some buffering, whats coming in will be ~10 times faster than whats being sent out and thus it would be SLOWER, due to the constant re-transmissons and dropped packets
Ffor this reason also, I had an idea that a 10 host multicast was the most efficient thing to do and so far its seemed about right - a single host takes 1.5 hours, but 10 take just under an hour to do. My guess was; less retransmissions because of the slower rate to each host.
If one PC was slowing the rest down, that would show in the log (have had this issue on mixed speed rooms). But it seems like they’re all hiccuping from the log
As for the switch things… its tricky because we have no way to access the switches. I work in the academic computing department at a University - normally, all departments are managed via a central ICT department but we have moved away from their systems and recently decided to use FOG. They keep all their switches under lock and key pretty much, so even physical access is limited. However, it seems that the particular switch (note, there are many switches in a stack) has no other hosts on it as far as I can tell. There might be a phone or a printer somewhere but I have no way of knowing. I’ll make a note to ask - communication face-to-face is difficult as they have to manage the infrastructure for the entire University - which involves multiple campuses. So there is a growing list of things to ask!
