Multicast is very slow

[tahitiju]

Hi,

I am facing a performance problem with FOG (v1.5.10).
I try to deploy an image to some PCs but performance is very low (2MB/min) using multicast.
For the purpose of the test, FOG server (1Gb NIC), clients (1Gb NIC) and dhcp are connected together using a Linksys Switch ( 16 x 1Gb ports / unmanaged) .
Using unicast, performance seems normal (~1,5GB/min deploying 14 PCs).

I tried things explained here to debug: multicasting
Testing 1 Client & Testing 2 Clients were OK, about transfering a small .fogsettings file
but i start to play with udp-sender to transfer a bigger (1GB sample ) file and :

• with 1 client, transfer is very fast
• with 2 clients, it was very slow (as like as my image deployment)

Where could be the problem ?

Thanks,

Julien

[george1421]

@tahitiju said in Multicast is very slow:

I try to deploy an image to some PCs but performance is very low (2MB/min) using multicast.
For the purpose of the test, FOG server (1Gb NIC), clients (1Gb NIC) and dhcp are connected together using a Linksys Switch ( 16 x 1Gb ports / unmanaged) .
Using unicast, performance seems normal (~1,5GB/min deploying 14 PCs).

A few things jump out at me with this. Multicasting can be very taxing on inexpensive network hardware. When you are getting 2MB/min how many systems are you multicasting to? Do you get the same performance when multcasting to 1 or 2 machines?

Multicast imaging moves at the speed of the slowest computer in the mcast group.

1,5GB/min is on the slow end of the scale. On a well managed 1GbE network, imaging to current hardware, I would expect about 6GB/min. If put your fog server on 10GbE network I would expect about 13GB/min. I’m only pointing it out that there might be something in your network infrastructure that is slowing things down.

A 1GbE network link will saturate with 3 simultaneous unicast deployments.

The FOG performance numbers displayed on the target computer are a composite number of how fast the fog server can pull the image off the disk, send it over the next, the client receiving the image, expanding it in memory and then writing it to disk. Any one component that is slow will impact the overall performance score.