Newbie MoCA questions

[krkaufman]

Clearly the X1s are communicating with MoCA 2.x despite the amplifier cutting off at 1002 MHz.

The X1 boxes are likely just standard or bonded MoCA 2.0, so don't have the same spectrum appetite of a MoCA 2.5 adapter ... and so are less effected by sub-optimal components.

 

[gneuner2]

Hi all,

So it turns out that the ECB6250s are working fine.

After consultation with ScreenBeam support (which was both responsive and helpful) it seems that the problem is high connection latency across the MoCA bridge interacting with system default TCP window sizes that are hopelessly inadequate.

To get decent performance across the bridge, TCP socket buffers (windows) have to be 1/2 MB or more for both transmit and receive. To achieve the full 1Gbps speed the bridge is capable of, the buffers have to be 2MB or larger. [On both sides of the bridge!]

This, of course, has its own problems. Although programs can use their own buffers or change the sizes of system allocated buffers, most programs will just use the defaults. If RSS (receive-side scaling) is enabled, the TCP stack can dynamically adjust the size of the receive buffer, but the size of the transmit buffer is fixed [unless the program itself adjusts it].

Increasing the default transmit buffer size affects every TCP socket in the system - not just ones dealing with high latency connections. Additionally, Windows has a single setting that is used for both transmit and receive ... Linux, at least, has separate settings.

So getting good performance means a lot more memory used. An active system has dozens of TCP sockets open all the time. Even having 32GB of RAM, I dislike the idea of every socket getting megabytes of buffer that many won't need [e.g., listen sockets just hand off incoming connections].

YMMV,
George

 

[tgl]

So it turns out that the ECB6250s are working fine.

After consultation with ScreenBeam support (which was both responsive and helpful) it seems that the problem is high connection latency across the MoCA bridge interacting with system default TCP window sizes that are hopelessly inadequate.

Wow, that is very interesting! (And at the same time, kind of a "doh!" moment.) I wonder if this effect explains the strange performance inconsistencies I've been seeing with my ECB7250s. I'd not thought to wonder how big the buffers are in the various units I've connected to them.