New Driver for Realtek RTL8111

[MacKonsti]

The name of interface isn't allocated by the driver but by the network stack itself. Is there something in the system which might cause the network stack to assume that there might be more ethernet interfaces? What do you get when you run "ifconfig -v" in Terminal?

 

Mieze

Hi Mieze, happy new year, it's the typical output, the board has only 1 LAN port anyway and it's detected properly (Gigabyte GA-H61N-USB3 running 10.10.5 latest):

 

CoolServer: Konsti$ ifconfig -v
lo0: flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> mtu 16384 index 1
eflags=10000000<SENDLIST>
options=3<RXCSUM,TXCSUM>
inet6 ::1 prefixlen 128
inet 127.0.0.1 netmask 0xff000000
inet6 fe80::1%lo0 prefixlen 64 scopeid 0x1
nd6 options=1<PERFORMNUD>
link quality: 100 (good)
gif0: flags=8010<POINTOPOINT,MULTICAST> mtu 1280 index 2
stf0: flags=0<> mtu 1280 index 3
en0: flags=8863<UP,BROADCAST,SMART,RUNNING,SIMPLEX,MULTICAST> mtu 1500 index 4
eflags=8c0<ACCEPT_RTADV,TXSTART,ARPLL>
options=6b<RXCSUM,TXCSUM,VLAN_HWTAGGING,TSO4,TSO6>
ether 74:d4:35:12:ae:3b
inet6 fe80::76d4:35ff:fe12:ae3b%en0 prefixlen 64 scopeid 0x4
inet 192.168.1.10 netmask 0xffffff00 broadcast 192.168.1.255
nd6 options=1<PERFORMNUD>
media: autoselect (1000baseT <full-duplex,flow-control>)
status: active
type: Ethernet
link quality: 100 (good)
scheduler: QFQ
link rate: 1.00 Gbps
CoolServer: Konsti$

Anyway, when I re-install OS X and use your v2.0.0 driver I will post more info, if the same thing happens.

 

Any approximation on the time of releasing your newer v2.0.1 driver for Realtek? Just to be sure my clean install uses v2.0.1 directly

 

Many thanks again for your valuable work!

[Mieze]

Any approximation on the time of releasing your newer v2.0.1 driver for Realtek? Just to be sure my clean install uses v2.0.1 directly

I'm currently working on an update for IntelMausiEthernet. As soon as it will be finished, I will resume work on the Realtek driver. As Realtek has published new Linux sources (V8.041) last month, just a few days after I published my last update, I decided to update the Linux sources again before going final with the new release just to make sure that the latest firmware and bug fixes are integrated in the next version.

 

Mieze

[R3TR0]

I have a small cosmetic suggestion. what if bypass hardcoded model name in kext, and take it from the device "model" data chunk? it would be great.

[Mieze]

I have a small cosmetic suggestion. what if bypass hardcoded model name in kext, and take it from the device "model" data chunk? it would be great.

Why? Personally I see no reason why you are proposing this change? By the way, source code can be found on GitHub and it shouldn't be too hard to do it yourself.

 

Mieze

[R3TR0]

For me personally, this is simply impossible )) anyway, thanks for the kext

it shouldn't be too hard

 

[dragonmel]

well I am at a loss.. I have issues and come to you all for some ideas..

 

I have a GA-EP45-ds3l fresh install (10.11.3) with clover boot loader using a patched DSDT and the latest RTL driver from here 

 

system boots fine.. seems to run fine.. internet loads.. I am getting just about book speeds for my charter broadband 60mbs down 3mbs up... 

 

I can connect from another computer and transfer file to and from the 10.11 install but if I connect FROM the fresh install to another computer I cant move files .. it locks up the finder

 

am I looking at a driver issue here or something else?

[Mieze]

well I am at a loss.. I have issues and come to you all for some ideas..

 

I have a GA-EP45-ds3l fresh install (10.11.3) with clover boot loader using a patched DSDT and the latest RTL driver from here 

 

system boots fine.. seems to run fine.. internet loads.. I am getting just about book speeds for my charter broadband 60mbs down 3mbs up... 

 

I can connect from another computer and transfer file to and from the 10.11 install but if I connect FROM the fresh install to another computer I cant move files .. it locks up the finder

 

am I looking at a driver issue here or something else?

The GA-EP45 boards have been reported to be troublemakers in the past because either there is a power management issue or a problem with interrupt handling but this isn't a driver issue.

 

Mieze

[dragonmel]

Thanks. The prev install was 10.10 with [url="http://www.insanelymac.com/forum/topic/279450-why-insanelymac-does-not-support-tonymacx86/"]#####[/url]/chameleons and it was working fine. I am at a loss right now

 

If you have any suggestions on where to look I would appreciate it

[dragonmel]

ok.. just an update...

 

@mieze good news and bad news... 

 

the good news is I have the system up and running fine.. the bad news is.. on this particular board.. it looks like it was your driver

 

I installed Linux2Mac's .9 release that I had been using up to 10.10 and it works like a charm.. on my gigE I am getting AFP transferes over 109MBS again... nice....

 

also it works for both the 10.10 and 10.11 installs I have on this board right now.. 10.10 still boots chameleon .. and the 10.11 boots clover

 

sorry....

[Mieze]

@mieze good news and bad news... 

 

the good news is I have the system up and running fine.. the bad news is.. on this particular board.. it looks like it was your driver

 

I installed Linux2Mac's .9 release that I had been using up to 10.10 and it works like a charm.. on my gigE I am getting AFP transferes over 109MBS again... nice....

 

also it works for both the 10.10 and 10.11 installs I have on this board right now.. 10.10 still boots chameleon .. and the 10.11 boots clover

No, there is no driver bug, it's your mainboard that's broken as it prevents the driver from servicing interrupts in time. Let me explain the problem a little bit in deep. Servicing interrupt requests is a task which runs at high priority and might easily have a devastating impact on system performance if the driver doesn't do proper interrupt moderation. In order to keep CPU load under control my driver delays  interrupts for received packets so that handling packets can be batched because chances are high that more packets will have been received when the interrupt handler actually runs. Therefore it is crucial that the platform allows interrupts to be serviced quickly when they occur and that's the point where you mainboard fails. I don't know if it's the interrupt management itself or a power management issue which causes this problem but it's definitely not the driver.

 

Lnx2Mac's driver is different, it's a piece of junk code which doesn't care for interrupt moderation or CPU load. It allows the NIC to generate tens of thousands of interrupts per second and eats up all available CPU time. That's the reason why it copes with your broken system very well.

 

Mieze

[dragonmel]

if there were a way that it could be fixed I would try it...

 

with your driver I get 10Mbs up instead of 120Mbs .. my plex server app is unable to serve content with your driver and I can stream 1080p native over wifi with the Linux2mac driver..

 

I understand that his might be a brute force method but it works

 

funny thing is that the slow UP speed is only if the transfer is initiated from the GA45... if its pulled from somewhere else the speed is better.. again.. weird.. 

 

I see no degradation of processor power.. and my scrubs on my 20TB ZFS pool are the same speed with his vs yours running in the background 

 

again.. if you have a suggestion on how to get yours to work on this board I would give it a shot.. but for now.. the linux kext has the system far more usable...

 

not busting on your work.. just stating that for this board.. this is evidently the way to go for now...

next up.. I have to upgrade my intel based boards to 10.11

 

D975xbx

D975xbx2

DX85so

 

I am not holding my breath on the original bad ax but do you do networking drivers for the intel chips as well?

 

sounds like you are the goto on the topic... and know far more about this than I do...

 

keep up the great work for the project...

[Mieze]

if there were a way that it could be fixed I would try it...

 

with your driver I get 10Mbs up instead of 120Mbs .. my plex server app is unable to serve content with your driver and I can stream 1080p native over wifi with the Linux2mac driver..

 

I understand that his might be a brute force method but it works

 

funny thing is that the slow UP speed is only if the transfer is initiated from the GA45... if its pulled from somewhere else the speed is better.. again.. weird.. 

 

I see no degradation of processor power.. and my scrubs on my 20TB ZFS pool are the same speed with his vs yours running in the background 

 

again.. if you have a suggestion on how to get yours to work on this board I would give it a shot.. but for now.. the linux kext has the system far more usable...

 

not busting on your work.. just stating that for this board.. this is evidently the way to go for now...

next up.. I have to upgrade my intel based boards to 10.11

 

D975xbx

D975xbx2

DX85so

 

I am not holding my breath on the original bad ax but do you do networking drivers for the intel chips as well?

 

sounds like you are the goto on the topic... and know far more about this than I do...

 

keep up the great work for the project...

The reason why things are worse when the transmission is initiated by the GA45 machine is TCP's congestion avoidance mechanism. In the Internet packet loss is caused by congested routers which are dropping packets and as TCP retransmits lost packets the situation would get worse and might even result in an unstable network. From the transmitters point of view congestion results in increasing and unpredictable packet roundtrip times. That's the reason why TCP implements a congestion avoidance mechanism which uses the packet roundtrip time and its standard deviation in order to estimate the connections throughput. This is the point where the trouble begins. The extremely long interrupt latency on your platform results in a delayed handling of received packets so that the TCP/IP stack gets confronted with increasing packet roundtrip times which also tend to vary and throttles the output rate as it assumes a congested network.

 

Of course you won't see a degradation of CPU performance on your machine but other users with "good" platforms have this problem using Lnx2Mac and there is a simple answer. As interrupts can't be retriggered until they have been served the extreme interrupt latency on your machine acts as some kind of interrupt rate throttling mechanism. Basically it's the same reason why a blind man is superior to sighted persons in a dark room.

 

Frankly, I don't know how to resolve this issue on your machine because I never had an affected system and I'm not quite sure if it can be resolved at all because it's a very old board and the interrupt architecture used to be a major weakness of the Intel platform in the past.

 

Mieze

[dragonmel]

thanks for the detailed explanation.. I still have a hard time wrapping my head around the idea that a file pulled from the machine would be handled differently than one pushed.. it is still packet traffic leaving the machine.

 

for now it will have to do, that machine has to serve in its media plex server role and it seems stable enough

 

are you familiar with any networking tricks or kexts that work best with the above intel boards?  some are saying that networking seems to work fine on some of them using clover...

 

I respect your expertise here.. please in no way think I am arguing with you.  As I said.. your knowledge here is far superior to mine.

 

I have been hackingtoshing back to the G4 days and if it were not for experts like you leading the way, I could not have gotten started.. so once again.. and I am sure you dont here it enough.. thanks for supporting the community

what may be making my situation worse, is I have a sil3132 and sil3124 card installed and the system will not even post if I use one specific 4 lane card that it has never liked.. sounds like IRQ conflicts.. which points to your interrupts theory there as well..

 

thanks again

[Mieze]

thanks for the detailed explanation.. I still have a hard time wrapping my head around the idea that a file pulled from the machine would be handled differently than one pushed.. it is still packet traffic leaving the machine.

The answer is simple because the congestion avoidance mechanism works on the transmitter's side, i.e. on the machine which initiated the connection. Years ago RehabMan reported a similar issue on a ProBook so that I took the time to do further research on performance relevant parameters for network drivers.

are you familiar with any networking tricks or kexts that work best with the above intel boards?  some are saying that networking seems to work fine on some of them using clover...

I don't have any experience with older hardware than the 2011 Sandy Bridge generation and due to the improved power management features of recent chipsets and CPUs I don't think that it's worth to consider recycling of older hardware in particular when you are planning to create a server. I developed this driver for my server, which was equipped with a RTL8111EV, 3 years ago. It worked flawlessly for 2 years with 24/7 uptime. Last year I decided to replace the machine with a new one based on an Asrock B85M-ITX (Atheros AR8171 NIC) and I'm happy with this machine too. Of course every NIC has it's advantages and disadvantages, there is no perfect solution and after all you'll have to keep in mind that there is no guarantee that a hackintosh will work as expected.

 

I respect your expertise here.. please in no way think I am arguing with you.  As I said.. your knowledge here is far superior to mine.

No problem! I appreciate when users are asking questions because they want to understand how things work as I don't see myself as a service provider but more as a teacher.

Edited January 30, 2016 by Mieze

[dragonmel]

No problem! I appreciate when users are asking questions because they want to understand how things work as I don't see myself as a service provider but more as a teacher.

 

​the better ones here are like that .. will take the time to explain it and walk you through it without holding their digital nose up!! hehe

 

I have gotten better at understanding this process but now that a new operating system is out every year and bootloads change just as quick I am lost again.. I have to make the switch to clover and am wrapping my head around how that works now.. more powerful but more complicated too..

[Mieze]

As already announced a few weeks ago here is a new development version of the driver which incorporates the following changes:

• Improved media selection and reporting. It is now possible to enable/disable flow control and/or EEE in System Preferences at runtime on chips which support it (chipset 14 and above).
• Updated the Linux sources to version 8.041.000.

Please keep in mind that this is a development version which may contain bugs. I have rewritten a lot of code and don't expect everything to work perfectly and I'm asking all users to test this driver thoroughly. Source code can be found on GitHub.

 

Good luck!

 

Mieze

RealtekRTL8111-V2.2.0d0.zip

[Mieze]

I have only one question, is there a chanse that I have now internet connection limited , like 1-5mb instead previous 20mb. Because I feel huge difference in a speed.. If yes is there some methods to fix this  problem?  Thanks!

Yes, there is and it's not a driver problem but a power management issue. The topic has been discussed in detail in this thread before (see above).

 

Mieze

[Multisaft]

Hi Guys, especially hello to Mieze,

 

I've found this thread as I know that insanelymac is my last resort for users that are diving deeper into problems that other topics in other communities are not able to resolve.

 

I am running a hackinotsh with the rtl8111c network card (Asus P6T-SE) on a clover configuration, yosemite 10.10.5, with as of today old network kext (I will look into details as soon as I'm back home). As soon as I'm back home I will edit this topic with every other information I have tried so far. 

 

The Problem: I'm running a 16TB EXT4 Open Media Vault NAS in my 1Gigabit Network (Netgear R7000 router, Cat5e Network cables) which has a Raid5 Performance of up to 550MB/s read and write. I'm using a lot of different shares with smb/cifs, afp, and nfs.

I've tested everything with the exact same topology which is:

NAS (OMV newest Linux Backport Kernel) <-> Cat5e cable <-> netgear r7000 router <-> Cat5e cable <-> Client pc - Hacintosh/Win10 Workstation

 

In Windows I see about 125MB/s (essentially capping my gigabit Ethernet connection) to the storage;

But with my Hackintosh I'm still stucked transferin huge files above 1GB with around 72MB/s read and 80MB/s write over smb2. I'm forcing yosemite to use smb2 with a /etc/nsmb.conf, CIFS/SMB1 is a little bit slower which is well known.

After all, that is the maxium performance I get after a lot of tuning the network stack of Yosemite within /etc/sysctl.conf. When you google, this is a well known fix without an exact configuration solving all problems. A lot of users posting that setting the delayed_ack=0 solved any issue maxing out the gigabit connections they use. I will post my configuration of /etc/sysctl.conf tonight as well as my OMV SMB socked configuraitons.

 

/etc/sysctl.conf:

# Increase some buffers

#net.inet.tcp.doautorcvbuf=0

#net.inet.tcp.doautosndbuf=0

net.inet.tcp.sendspace=4194304

net.inet.tcp.recvspace=4194304

net.inet.tcp.maxseg_unacked=32

net.inet.tcp.delayed_ack=0

net.inet.tcp.win_scale_factor=8

#net.inet.tcp.rfc1323=1

 

# increase OSX TCP autotuning maximums

net.inet.tcp.autorcvbufmax=33554432

net.inet.tcp.autosndbufmax=33554432

 

 

/etc/nsmb.conf:

[default]

smb_neg=smb2_only

 

 

SMB Socked configuration in OMV: 

max protocol = SMB2

max xmit = 65535

socket options = TCP_NODELAY IPTOS_LOWDELAY SO_SNDBUF=65535 SO_RCVBUF=65535

read raw = yes

write raw = yes

max connections = 65535

max open files = 65535

 

 

This is a well known source for network stack tuning under unix/linux/os x: https://rolande.wordpress.com/2014/05/17/performance-tuning-the-network-stack-on-mac-os-x-part-2/

 

 

Does anyone in this forum ever encountered a situation as described above? 

I hope that the new driver you are supporting us with Mieze will solve all my problems (I will test it tonight).

[Mrengles]

Hi Guys, especially hello to Mieze,

 

I've found this thread as I know that insanelymac is my last resort for users that are diving deeper into problems that other topics in other communities are not able to resolve.

 

I am running a hackinotsh with the rtl8111c network card (Asus P6T-SE) on a clover configuration, yosemite 10.10.5, with as of today old network kext (I will look into details as soon as I'm back home). As soon as I'm back home I will edit this topic with every other information I have tried so far. 

 

The Problem: I'm running a 16TB EXT4 Open Media Vault NAS in my 1Gigabit Network (Netgear R7000 router, Cat5e Network cables) which has a Raid5 Performance of up to 550MB/s read and write. I'm using a lot of different shares with smb/cifs, afp, and nfs.

I've tested everything with the exact same topology which is:

NAS (OMV newest Linux Backport Kernel) <-> Cat5e cable <-> netgear r7000 router <-> Cat5e cable <-> Client pc - Hacintosh/Win10 Workstation

 

In Windows I see about 125MB/s (essentially capping my gigabit Ethernet connection) to the storage;

But with my Hackintosh I'm still stucked transferin huge files above 1GB with around 72MB/s read and 80MB/s write over smb2. I'm forcing yosemite to use smb2 with a /etc/nsmb.conf, CIFS/SMB1 is a little bit slower which is well known.

After all, that is the maxium performance I get after a lot of tuning the network stack of Yosemite within /etc/sysctl.conf. When you google, this is a well known fix without an exact configuration solving all problems. A lot of users posting that setting the delayed_ack=0 solved any issue maxing out the gigabit connections they use. I will post my configuration of /etc/sysctl.conf tonight as well as my OMV SMB socked configuraitons.

 

/etc/sysctl.conf:

 

/etc/nsmb.conf:

 

SMB Socked configuration in OMV: 

socket options = TCP_NODELAY IPTOS_LOWDELAY SO_RCVBUF=65536 SO_SNDBUF=65536

 

 

This is a well known source for network stack tuning under unix/linux/os x: https://rolande.wordpress.com/2014/05/17/performance-tuning-the-network-stack-on-mac-os-x-part-2/

 

 

Does anyone in this forum ever encountered a situation as described above? 

I hope that the new driver you are supporting us with Mieze will solve all my problems (I will test it tonight).

 

@Multistaf & Mieze

 

Sorry for going off topic, but I though this would benefit Multistaf with his overall NAS transfer rates. The link below should help your R7000/NAS combo quite-a-bit. Keep in mind it will do nothing for your driver performance tuning or answer any of your questions.

 

https://github.com/RMerl/asuswrt-merlin/wiki/Link-Aggregation  * I've tried this with other R7000 open source firmwares like DD-WRT/Tomato without success.

 

Good luck!

 

Again sorry Mieze.

 

Regards,

 

Robert aka Mrengles

[Mieze]

Hi Guys, especially hello to Mieze,

 

I've found this thread as I know that insanelymac is my last resort for users that are diving deeper into problems that other topics in other communities are not able to resolve.

 

I am running a hackinotsh with the rtl8111c network card (Asus P6T-SE) on a clover configuration, yosemite 10.10.5, with as of today old network kext (I will look into details as soon as I'm back home). As soon as I'm back home I will edit this topic with every other information I have tried so far. 

 

The Problem: I'm running a 16TB EXT4 Open Media Vault NAS in my 1Gigabit Network (Netgear R7000 router, Cat5e Network cables) which has a Raid5 Performance of up to 550MB/s read and write. I'm using a lot of different shares with smb/cifs, afp, and nfs.

I've tested everything with the exact same topology which is:

NAS (OMV newest Linux Backport Kernel) <-> Cat5e cable <-> netgear r7000 router <-> Cat5e cable <-> Client pc - Hacintosh/Win10 Workstation

 

In Windows I see about 125MB/s (essentially capping my gigabit Ethernet connection) to the storage;

But with my Hackintosh I'm still stucked transferin huge files above 1GB with around 72MB/s read and 80MB/s write over smb2. I'm forcing yosemite to use smb2 with a /etc/nsmb.conf, CIFS/SMB1 is a little bit slower which is well known.

After all, that is the maxium performance I get after a lot of tuning the network stack of Yosemite within /etc/sysctl.conf. When you google, this is a well known fix without an exact configuration solving all problems. A lot of users posting that setting the delayed_ack=0 solved any issue maxing out the gigabit connections they use. I will post my configuration of /etc/sysctl.conf tonight as well as my OMV SMB socked configuraitons.

I'm almost 100% sure that this is a power management issue and not a driver problem. With El Capitan Apple greatly increased the time the CPU cores spend in a deep sleep state and this may impact I/O performance because it takes some time for the CPU to exit the low power sleep states which also means that execution of interrupt service routines will be delayed when the CPU has to wake up in order to service an interrupt request from the NIC. This means that you will have to change the power management settings (patch the SSDT to discourage the CPU from entering a deep C-state) for better network performance.

 

There might be another approach: Apple introduced a polled mode in their new network driver interface. Basically it works much like Linux's NAPI where the network stack is allowed to dynamically switch the driver to a polled mode under load, i.e. the network stack polls the driver for received packets periodically instead of letting the driver cause an interrupt when it has new packets to process. For further information please see: https://en.wikipedia.org/wiki/New_API

 

The polled mode has two advantages over interrupt driven mode under heavy load:

• It saves the processing overhead induced by interrupt handling.
• It improves cooperation of the driver with CPU power management because a scheduled poll task can be taken into account when considering to put a core into a deep C-state whereas the point in time when the next interrupt will occur is unpredictable.

I'm currently trying to implement polled mode for the driver and will publish a test version as soon as possible.

 

Mieze

[Multisaft]

Hi guys,

thanks for your feedback. I've posted my config above. Still I dont think that it is a problem of needed link aggregation (this will be the next step in my build process) cause my win10 workstation does cultivate every possible ethernet connection whereas the os x clients in my network are not able to use full bandwith. 

 

I'm going to install your last driver mieze. I will be back in some hours for a feedback.

 

Feedback about the 2.00 Realtek driver: 

 

I'm still hovering around the 80mb/s read and now I see write speeds to my raid5 of about 1-2MB/s but I'm running a huge rsync job at the moment which will take around 20 more hours to complete. I will test write speeds over smb2 tomorrow again.

 

so I think that doesn't resolve my issues with poor gigabit connections.

 

In addition. AFP performance is even worth transfering large files using the path finder (write around 20mb/s when there is no rsync job running).

When I use iperf and iperf3 for network performance tests, I see around 550mbps speed, only half the possible gigabit bandwith. Using a hdd test like blackmagic disk speed test I get full 125MB/s throuput over smb connection. what is wrong with this yosemite configuration. I'm getting mad. 

[Mieze]

In addition. AFP performance is even worth transfering large files using the path finder (write around 20mb/s when there is no rsync job running).

When I use iperf and iperf3 for network performance tests, I see around 550mbps speed, only half the possible gigabit bandwith. Using a hdd test like blackmagic disk speed test I get full 125MB/s throuput over smb connection. what is wrong with this yosemite configuration. I'm getting mad. 

It's a power management issue! The less load you have, the worse it gets because the CPU is dozing.

 

Mieze

[RehabMan]

Hi guys,

thanks for your feedback. I've posted my config above. Still I dont think that it is a problem of needed link aggregation (this will be the next step in my build process) cause my win10 workstation does cultivate every possible ethernet connection whereas the os x clients in my network are not able to use full bandwith. 

 

I'm going to install your last driver mieze. I will be back in some hours for a feedback.

 

Feedback about the 2.00 Realtek driver: 

 

I'm still hovering around the 80mb/s read and now I see write speeds to my raid5 of about 1-2MB/s but I'm running a huge rsync job at the moment which will take around 20 more hours to complete. I will test write speeds over smb2 tomorrow again.

 

so I think that doesn't resolve my issues with poor gigabit connections.

 

In addition. AFP performance is even worth transfering large files using the path finder (write around 20mb/s when there is no rsync job running).

When I use iperf and iperf3 for network performance tests, I see around 550mbps speed, only half the possible gigabit bandwith. Using a hdd test like blackmagic disk speed test I get full 125MB/s throuput over smb connection. what is wrong with this yosemite configuration. I'm getting mad.

Just as a quick test, you might try my fork: https://github.com/RehabMan/OS-X-Realtek-Network

 

It deals with interrupts in a slightly different way...

[Multisaft]

guys thanks again. I'm sitting in Germany, i will try it tomorrow, rsync is still migrating another NAS ({censored} zyxel nsa325v2 which is used  as a remote backup server and family storage) at the moment at just around 20mb/s (100% cpu load on the zyxel NAS... zZzzz). 

I will give a feedback tomorrow.

[Multisaft]

ok, 

 

the rsync is done. both of your forks are having some issues on my configuration. I'm seeing write speeds at around 1-2 MB/s over smb2 and a bandwith from iperf at around 12 Mbits/sec. 

with my old RTL81xx.kext I get around 60 to 70 MB/s write speeds to the network share (smb2) and iperf bandwith at around 511 Mbits/sec (still not duplex 1gigabit). and what I dont get: with my RTL81xx.kext I see around 104 MB/s write and 90MB/s read speed on my smb2 network share using the black magic disk speed test.... what is happening here.

 

In addition another test. I can read and write from two different shares on the same network storage - mounted with smb2 - capping my gigabit connection (55MB/s read, 55MB/s write).