High Sierra native support for 10Gb ethernet

[ydeng]

(Sonnet Solo10G PCIe in a Thunderbolt enclosure)



`IOBuiltin` is `False`, if that's what you mean.

View attachment 773208



How would I search for that?

Two levels up in IOService plane. DSDT/SSDT is in IOACPIPlane. The screen shots are from onboard AQC.

iMacPro does not have built-in property in DSDT:

 

[Kris Kelvin]

@ydeng Can't seem to find the corresponding node in IOACPIPlane...

 

[ydeng]

@ydeng Can't seem to find the corresponding node in IOACPIPlane...

View attachment 773301

Thanks for trying. Then there is no DSDT/SSDT for the device.
[doublepost=1532815313][/doublepost]

A short test yielded the following numbers:

Setup:
System 1: MacPro5,1 ⇄ Chelsio T520-CR (slot 2) ⇄ Mikrotik S+RJ10
System 2: MacPro6,1 ⇄ ThunderTek/PX (Thunderbolt 1 enclosure with small fan) ⇄ Sonnet Solo10G PCIe
Both systems running High Sierra 10.13.6. Using a RAM disk on the MacPro5,1 to make sure there's enough IO bandwidth to the "disk". Connected by a Cat6 cable (no switch inbetween). MTU 1500 (no Jumbo frames).

Temperatures:
Ambient: 27°C
Chelsio adapter: heatsink up to 52°C
Sonnet adapter: heatsink up to 72°C
The MacPro has better airflow than the Thunderbolt enclosure. When mounted in the MacPro5,1, the Sonnet did only get up to 60°C.

Speed:
iPerf: 5-6 GBit/s
SMB: 500MB/s (default config; signing enabled)

That's good, but only about half of the theoretical speed. Not sure where the bottleneck is. Thunderbolt 1? Negotiated PCIe speed? One of the adapters? Drivers? MTU? CPU? Cable? Something else?
Still, it means that you can pop a Solo10G PCIe adapter into a MacPro5,1 and get 500MB/s out of the box (if the connected device supports it), and that's quite nice.

HackinTosh:

iperf -c is about 9.7G on AQC107
iperf -s is only 4.7G

Linux both -c and -s are 9.7G.

 

[d5aqoëp]

Good news! I've received my Sonnet Solo10G PCIe card and it works as advertised. Plug and play in both a MacPro5,1 and a Thunderbolt enclosure (FirmTek ThunderTek/PX) under macOS 10.13.5.

Now I need a second device to actually test 10G troughput.

View attachment 770053 View attachment 770054 View attachment 770055 View attachment 770056 View attachment 770057 View attachment 770058
PS: Sorry for the german screenshots. Didn't realize it at the time.

Can you test if WOL Wake on Lan is working in macOS and in Windows?

 

[mikeboss]

A short test yielded the following numbers:

Setup:
System 1: MacPro5,1 ⇄ Chelsio T520-CR (slot 2) ⇄ Mikrotik S+RJ10
System 2: MacPro6,1 ⇄ ThunderTek/PX (Thunderbolt 1 enclosure with small fan) ⇄ Sonnet Solo10G PCIe
Both systems running High Sierra 10.13.6. Using a RAM disk on the MacPro5,1 to make sure there's enough IO bandwidth to the "disk". Connected by a Cat6 cable (no switch inbetween). MTU 1500 (no Jumbo frames).

Temperatures:
Ambient: 27°C
Chelsio adapter: heatsink up to 52°C
Sonnet adapter: heatsink up to 72°C
The MacPro has better airflow than the Thunderbolt enclosure. When mounted in the MacPro5,1, the Sonnet did only get up to 60°C.

Speed:
iPerf: 5-6 GBit/s
SMB: 500MB/s (default config; signing enabled)

That's good, but only about half of the theoretical speed. Not sure where the bottleneck is. Thunderbolt 1? Negotiated PCIe speed? One of the adapters? Drivers? MTU? CPU? Cable? Something else?
Still, it means that you can pop a Solo10G PCIe adapter into a MacPro5,1 and get 500MB/s out of the box (if the connected device supports it), and that's quite nice.

try MTU 9000
try NFS instead of SMB/CIFS -> https://www.bresink.com/osx/NFSManager.html
cable should be Cat. 6A
according to your screenshot, the link on the PCIe is only 4x 2.5 GT/s (should be 4x 5.0 GT/s IMHO)
[doublepost=1547112412][/doublepost]also, AFAIK, the first batch of Sonnet 10GbE PCIe cards had some problems.
Rev. A cards: serial starts with an "A". rev. B cards: serial starts with a "B".
the serial # of your card starts with an "A"...

 

[PTuser]

Hi - Things have moved on since this thread started - Im on 10.13.6 with a Hackintosh and a Mac Pro. I bought a Syba 1 10G PCIe card for both machines using a direct P to P cable link. It has the ACQ107 chipset. Both machines see and recognise this card and Network sys prefs sees it as a 10gig connection. Im getting very poor transfer speeds. emailed support who suggest Kernel patching the sysctl.conf file. Is this still valid for 10.13.6? Tried a variety of cables...to no avail......Im stuck....I can return the cards but would hate to do so if theres something I could do to get a half decent speed

 

[nick314]

Hi - Things have moved on since this thread started - Im on 10.13.6 with a Hackintosh and a Mac Pro. I bought a Syba 1 10G PCIe card for both machines using a direct P to P cable link. It has the ACQ107 chipset. Both machines see and recognise this card and Network sys prefs sees it as a 10gig connection. Im getting very poor transfer speeds. emailed support who suggest Kernel patching the sysctl.conf file. Is this still valid for 10.13.6? Tried a variety of cables...to no avail......Im stuck....I can return the cards but would hate to do so if theres something I could do to get a half decent speed

Having done a lot of tinkering, I would highly recommend that your return the card and buy this one which has native support in High Sierra and later. https://www.sonnettech.com/product/solo10g-pcie-card.html

Just make sure you add it to a pci-e 2.0/3.0 x16 slot for max speeds.

 

[killawat]

Hi - Things have moved on since this thread started - Im on 10.13.6 with a Hackintosh and a Mac Pro. I bought a Syba 1 10G PCIe card for both machines using a direct P to P cable link. It has the ACQ107 chipset. Both machines see and recognise this card and Network sys prefs sees it as a 10gig connection. Im getting very poor transfer speeds. emailed support who suggest Kernel patching the sysctl.conf file. Is this still valid for 10.13.6? Tried a variety of cables...to no avail......Im stuck....I can return the cards but would hate to do so if theres something I could do to get a half decent speed

use iperf (direct link: iPerf 3.1.3) on both machines, on the server run
iperf -s -p 9000
for the server
iperf -c 1.2.3.4 -p 9000

 

[PTuser]

Hi Killawat
Thank you for replying and the info. Its a direct link - which is two way - not sure which qualifies as the server. Just so I understand I have downloaded the iperf you suggested. Is it a terminal command? and will it affect my separate onboard 1g connection, which is used for router based internet etc? Apologies for my limited knowledge ....

 

[Kris Kelvin]

Its a direct link - which is two way - not sure which qualifies as the server.

Either one; in case if iperf, it doesn't matter.

Just so I understand I have downloaded the iperf you suggested. Is it a terminal command?

Yes, it's a terminal command.
There's also a (paid) GUI version in the Mac App Store: https://apps.apple.com/ch/app/iperfutil/id1033018345?mt=12

and will it affect my separate onboard 1g connection, which is used for router based internet etc?

No, it will not affect your other network connections.

However – if you have multiple connections, your Mac will choose the topmost one. So maybe your Macs are not using the fast link. Do you mind sharing screenshots from System Preferences?

 

[Skyfokker]

Hey guys my first post here and really need some help. I have 3 Mac pros linked with aquaria aqc107 pcie NIC to a netgear XS508m unmanaged switch. Cat 6A cables with one of them (server) running twin SSD RAID 0 capable of 1000mb/s. Running Blackmagic speed tests over network I only get read write speeds around 300mb/s. Similar results on SMB or AFP.

OSX 10.13.6 and even tried a fresh os installation with nothing extra installed but still same results. My MTU remains at 1500 ( switching it to 9000 slowed down my wireless MBPs on the rest of the network )

What am I doing wrong? I’m about to go mad. IPV6 should be turned off?

 

[Kris Kelvin]

Running Blackmagic speed tests over network I only get read write speeds around 300mb/s.

With that, you're testing a long chain of things at the same time (drives, RAID software, filesystem, SMB client, SMB server, Blackmagic Disk Speed Test App).

To find out more about just your networking situation: What are your iperf results?

 

[Skyfokker]

With that, you're testing a long chain of things at the same time (drives, RAID software, filesystem, SMB client, SMB server, Blackmagic Disk Speed Test App).

To find out more about just your networking situation: What are your iperf results?

I haven’t tried that. Does that provide valid info? My cards are in slot 3 of my 5,1 Mac pros

 

[PTuser]

Hi Kris
Thank you for the info. Any test Ive done on the 10g side, Ive disabled the 1G connection on both in system prefs before testing

Thank you for taking an interest - Ill post the screenshots shortly

 

[pastrychef]

Hi - Things have moved on since this thread started - Im on 10.13.6 with a Hackintosh and a Mac Pro. I bought a Syba 1 10G PCIe card for both machines using a direct P to P cable link. It has the ACQ107 chipset. Both machines see and recognise this card and Network sys prefs sees it as a 10gig connection. Im getting very poor transfer speeds. emailed support who suggest Kernel patching the sysctl.conf file. Is this still valid for 10.13.6? Tried a variety of cables...to no avail......Im stuck....I can return the cards but would hate to do so if theres something I could do to get a half decent speed

I use a Syba 10GBase-T everyday on my mackintosh and never had to do any patching to sysctl.conf. The card uses native macOS drivers and works OOB.

I've seen others report slow speeds when using AQC107 based cards in older Mac Pros. Might be a PCI-e 2.0 thing or limited number of lanes. I suggest trying a different PCI-e slot.

 

[PTuser]

Either one; in case if iperf, it doesn't matter.



Yes, it's a terminal command.
There's also a (paid) GUI version in the Mac App Store: https://apps.apple.com/ch/app/iperfutil/id1033018345?mt=12



No, it will not affect your other network connections.

However – if you have multiple connections, your Mac will choose the topmost one. So maybe your Macs are not using the fast link. Do you mind sharing screenshots from System Preferences?

Hi Kris - Typing both those commands in terminal are giving me command not found on each machine?

 

[PTuser]

I use a Syba 10GBase-T everyday on my mackintosh and never had to do any patching to sysctl.conf. The card uses native macOS drivers and works OOB.

I've seen others report slow speeds when using AQC107 based cards in older Mac Pros. Might be a PCI-e 2.0 thing or limited number of lanes. I suggest trying a different PCI-e slot.

Hi - Its in an 8 way lane on my Hackintosh and a 4 way on the Mac Pro - The hackintosh is full but there are options on the Mac Pro. what kinds of speeds are you getting? would you mind sharing your sys prefs settings for it? Im just going direct between the cards
[doublepost=1560871586][/doublepost]Ipref is running but I can't get kilowatts commands to run - command not found
[doublepost=1560871971][/doublepost]Hi Kris - Here are the requested sys prefs -

 

[PTuser]

ok so I bought the commercial iperf in the end- the host 10gig connection showing 37.21gbits/sec. Slave is showing 28.00 Gbits/sec,on the 10g connection. The 1gb connection is showing 36.87Gbits/sc Host 27.85 Gbits/sec. When testing the 10g the 1g is made inactive on each comp and vice versa - I cannot understand these results at all....

 

[killawat]

ok so I bought the commercial iperf in the end- the host 10gig connection showing 37.21gbits/sec. Slave is showing 28.00 Gbits/sec,on the 10g connection. The 1gb connection is showing 36.87Gbits/sc Host 27.85 Gbits/sec. When testing the 10g the 1g is made inactive on each comp and vice versa - I cannot understand these results at all....

What are your cable specs? Are you using cat 6/cat 7? Long run? Short run?
Also turn off jumbo frames for now to see if you get anything different.

 

[PTuser]

What are your cable specs? Are you using cat 6/cat 7? Long run? Short run?
Also turn off jumbo frames for now to see if you get anything different.

Iv tried 3 cables - Two Cat6a - one worked but lowered scores - the other didn't - Best results are from a CAT 8 - all less than a metre.twisted pair copper as per the specs of the card
I moved the card in the Mac Pro to a x16 slot. Remote - local is 36.56Gbits/sec but slightly lower in that slot from local to remote.

Tried turning off jumbo packets - no difference in results

 

[Kris Kelvin]

the host 10gig connection showing 37.21gbits/sec. Slave is showing 28.00 Gbits/sec,on the 10g connection.

That looks like connections to localhost, but you want to test the connection the the other machine. Obviously, the network connection can't be faster than 10Gbits/sec.

Try again like this:

1. Launch iperfUtil on Mac 1 and make sure the bottom left switch ("Local server") is enabled
2. Launch iperfUtil on Mac 2, enter the IP address of Mac 1 in the field "IP address" of section "Remote server" and click "Start"
3. By default, this configuration will test transfer speeds from Mac 1 to Mac 2. You can change that with the dropdown "Mode".

 

[PTuser]

Hi Kris
so on the slave (Mac 2) using your instructions bandwidth local to remote was 6.34Gb/s - Remote to local was 1.34gb/sec - on second test remote to local is now 2.61gb/sec....Ive left it going for 30 mins - its now up to 3.05 Gb/sec

 

[Kris Kelvin]

so on the slave (Mac 2) using your instructions bandwidth local to remote was 6.34Gb/s - Remote to local was 1.34gb/sec

That's definitely not enough and indicates a problem with the network connection itself.

Can you check the error rates in "Network Utility.app"? Switch to the first tab ("Info"). There is a dropdown that lists the different network interfaces. Select the 10G interface, run an iperf test again and check the number of "Send Errors" and "Recv Errors". With a good ethernet connection, both these counters should be 0.

 

[PTuser]

Can you check the error rates in "Network Utility.app"? Switch to the first tab ("Info"). There is a dropdown that lists the different network interfaces. Select the 10G interface, run an iperf test again and check the number of "Send Errors" and "Recv Errors". With a good ethernet connection, both these counters should be 0.[/QUOTE]

Heres are both screenshots with all tests from mac2

Heres a link to what Syba sent me yesterday - most of it refers to pre 10.13.6 though.....https://support.studionetworksoluti...10GbE-recommendations-and-troubleshooting-v2-

 

[Kris Kelvin]

Heres are both screenshots with all tests from mac2

Hm, both error counters are 0, which means that the connection is good (so in my opinion the cable and shielding are fine).
Since testing the localhost with iperf yielded troughput far above 10gbit/s, the CPU shouldn't be the problem either.

Is it possible that the cards run too hot? The AQC107 chipset tends to throttle at high temperatures.
Or do you happen to have some firewall software running that might be slowing the connection down?