MOCA 2.5 - Asymmetrical speed

[45263tec]

Posted this on reddit but i don't expect much from there for reasons i won't get into.

I have 2 new GoCoax MA2500D MOCA adapters with 2.5GBE switches, cat 6 cable and 2 machines running RTL8125b 2.5GBE network cards.

The problem I'm having is that transfers to the file server write at close to 210MB/s (not great, but good enough) but the read speed is terrible at only 50MB/s.

I have turned off EEE on both machines, MA2500D updated to latest firmware. The MOCA splitters in use are wideband and explicitly MOCA compatible. One holland 3-way moca splitter with a POE filter outside the house, one "starburst" moca compatible 2-way indoors.

I don't expect mega speeds when reading and writing are occurring at the same time, but what could the issue be?

Here's what I tried so far.

• LOOOONG ethernet cable wiring both machines together = full 2.5gbps speed in both directions.
• LOOOONG ethernet cable connecting both 2.5Gb switches together = full 2.5gbps speed in both direcitons
• Both MOCA adapters plugged directly into each machine, same problem.
• I powered off my cable modem, same problem.

PHY rates in the UI are perfect:
PHY rate (0-1): 3523
PHY rate (1-0): 3531
PHY rate (0-0): 696
PHY rate (1-1): 693

No bad packets

Iperf3 bears out the issue. Using 5 threads with (-P 5):

Client -> File server 2.34 Gbits per second
File server -> Client 556 Mbits per second.

This basically matches what I am getting manually transferring files or running AJA.

Client -> Fileserver: 215 MBytes per second (write)
Fileserver -> Client: 55 MBytes per second (read)


I have seen a couple posts discussing this topic but no one seems to have any answers. Hoping someone has a fix.

 

[degrub]

Your moca network seems fine given the sync rate. Your issue is likely the server read/write or client r/w or protocol used (SMB ?) . The iperf tests should be running from memory only, so disk access would not be in play . Try -p 10 and run it in both directions using reverse switch setting.

When you write 2.5 Gb/sec on the long ethernet cable test - that is the link rate shown for the interface ?
If so, that has nothing to do with the transfer rate except to set an upper bound.

 

[45263tec]

Yes this is using SMB. What bothers me is that with both machines connected via ethernet (no MOCA) the transfer rate is pegged at 290-300 something megabytes a second which is what the drives in the file server can do.

-P 10 shows good speed, close to 2.34gbps with a fot of fluctuation but frankly this is not of any use if the file transfers themselves are extremely slow in day to day work. For example, 50MB/s is just not enough to use files live off the file server.

 

[degrub]

That 2.34 Gb/s is a saturated 2.5 link.
if you need real time access in the file, then use direct attached storage as that is the cheapest way to go. TB4 or 3 or maybe a local nvme array if you have enough pci lanes and bandwidth left available.

see the diy nas sub forum.

 

[thewifiatemyhomework]

I am having similar results…2.3 upload but only 800-1200 download in reverse mode via iperf3 even with multiple streams. Worked fine in a controlled bench test using minimal coaxial cable length…could be the coaxial cabling inside the walls…

 

[tgl]

There are similar reports in other threads, eg here