This same box, Slackware 13.1 installed and updated:

Copies from a Win7 system to a simple unprotected samba share stabilize at ~85MB/s according to Explorer. This is with kernel 2.6.33.4, r8169 module 2.3LK-NAPI. That seems to duplicate reports of better experience with this nic using the latest kernel & driver. No idea yet how it'll build/work against 5.06a's kernel.

Interesting thread. I am sure my proposed motherboard for myunraid server uses a similar NIC. I will check the model and post back. Is it not recommended to run an Intel Pro anyway for best speeds with unraid boxes? Are they pricey? I take it they are PCI-e now? EDIT: Just had a quick look on scan.co.uk and they have an Intel Pro PCI for £20 and a PCI-e dual port for £120. I take it the PCI gigabit intel pro is still ok to use and no reason to get the PCI-e dual port one for a lot more money? (unless you need dual port which i don't)

They do get the most attention. They've pretty much been the standard since taking over from 3C5 cards, what, 20yrs ago.

 

There's also a PCIE single-port card for about the same price as the PCI, but PCI can handle a single Gb port so long as nothing else is fighting for attention.

So is that potentially the fix? To get a PCI(E) NIC card? Would Intel work with an AMD mobo, or does it require an Intel chipset?

 

Ultimately, I guess we should wait to see if they can update the NIC drivers in unRAID.  Or if it's possible for us to do it manually.

An Intel nic will work fine in an AMD chipset... Not like it will try and recompile the BIOS.

The HP card I am using is a Broadcom chipset.

 

Shawn

 

So is that potentially the fix? To get a PCI(E) NIC card? Would Intel work with an AMD mobo, or does it require an Intel chipset?

 

Ultimately, I guess we should wait to see if they can update the NIC drivers in unRAID.  Or if it's possible for us to do it manually.

 

Whenever I have read on forums people talking of backup/transfer speeds on their home lan, more often than not the most popular recommendation is to reduce the bottleneck of realtek or other onboard ethernet, and switch to an Intel pro dedicated card. Whether or not this is effective I cannot say from first hand experience, but there seems to be a fair following for using them as they have better throughput. i.e. Yes gigabit ethernet in theory is capable of 125 MB/s, but you will never get this in reality. Intel Pro can make it be nearer to that number. However, I will personally be seeing what I get out oy my realtek before potentially buying an Intel. Every little helps I guess.

 

Yes gigabit ethernet in theory is capable of 125 MB/s, but you will never get this in reality. Intel Pro can make it be nearer to that number.

 

Indeed, iperf shows that I can achieve 114MB/s between two machines with Intel Pro embedded network controllers, with two consumer grade switches in the path.

I was maxing out at about 65 MB/s and then the connection died.  My WiFi to the APE was still alive, but the Ethernet port died on my Mac (I'm connected on both, just changed the Preference Order to WiFi being first after it failed).  I'm wondering if it's the program (ForkLift) that I was using to transfer (because it actually shows you your speeds--unlike Finder).  So I'm not sure if the Ethernet connection on the APE actually dies, or if ForkLift is responsible for killing the Ethernet on my Mac, or if the Realtek on the Server is also dying...

After a bunch of fruitless messing about with kernel parameters I ran across posts describing video and nic performance problems with this board under Windows. The fix was to disable various power saving features in the Windows drivers. Sounded relevant and familiar (old lessons learned again), so I disabled everything I could find related to power saving in the EFI gui. Short version (all I have so far) is things seem to be working. No IRQ Disabled messages since and workstation to unRAID copies stabilize ~65MB/s (no parity, per Win7 progress). Some irony here given my original intent of power savings, but then I haven't checked power consumption differences after the settings or which settings are really important. Hopefully I'll have time to play with it more this weekend. Meantime it's rebuilding parity for another 10hrs.

So where did you do this? Under BIOS? This happened again using Finder, but I couldn't find anything obvious in the Mac, APE, or unRAID logs.

So where did you do this? Under BIOS? This happened again using Finder, but I couldn't find anything obvious in the Mac, APE, or unRAID logs.

 

Whoops, sorry. Kind of late.

 

What I was trying to do was disable power-saving features that could result in extra effort required of the system, more latency, and less capability. That meant disabling anything I could find in the board's EFI gui (not much) and similar kernel options e.g. pcie_aspm=off. To me the problem sounds different than yours, though fundamentally they may be very similar. Problems at the edge of performance usually are.

 

Don't rule out your switch/router. They can throw in some extremely annoying side-effects. If you don't step back after awhile and test things against a different device you can easily waste time. On to my next post...

I brought my server to my friends' house to load up on some data, and this happened again.  It took down his whole network (just like at my house)

 

Had to do a hard reset on the router, then I just transferred one by one, but that is very annoying.  There must be a way to fix this, without purchasing a new NIC card..

 

He also has a different router than me.  I have an APE, he has a Dlink Xtreme Gaming router or something like that.  So, what else? And if I got an Intel NIC, would PCI be fine? I only have a PCI-E (x1), 2x PCI, and a PCI-E (x16)--which I heard are mostly for video cards? Filling up the x1 PCI-E with a SATA card, and may need more, so should a normal PCI port be adequate for an Intel gigabit NIC?

... should a normal PCI port be adequate for an Intel gigabit NIC?

Yes.  To fix my 8111C problems I got a PCI card because I didn't want to sacrifice a PCI-e slot and it works perfectly.

Ugh this is annoying.  Onboard NIC should be fine, do we think its a driver issue then?  How can we upgrade the drive manually? I'd like to try that before buying a card, if possible...

For now, since there appear to be several causes or twists to the Realtek problem, for me it's best to go the easy route and add a card. I really didn't want to but better people are working on it. I expect we'll see later releases of the driver able to contend with our boards. Until Realtek throws in more versions. Dilbert R&D: "If they've worked around all the problems we aren't creating enough of them."

 

Yes, PCI can just handle GbE. Don't use the other PCI slot, there isn't much headroom. I bought the usual Intel Pro GT - PWLA8391GTBLK - $30. You could probably use a cheaper one but chipsets and variations make it luck of the draw.

Onboard NIC should be fine,

 

Yes, it should be ... but in my experience of Realtek, stemming back over 15 years, Realtek has always given me headaches - except, for some reason, the Netgear GA311 using the 8169.

 

As I've stated before, I always avoid motherboards with Realtek chips (just troublesome) and Atheros chips (poor driver support).

 

Just buy yourself a decent nic card (PCI is fine, assuming that there is no other heavy user of the bus) - either the GA311 or an Intel.

Does it seem like my problem is similar? Would the card take down the whole network? What happens, data gets backed up because of the NIC and supposedly brings down the network?  Will this new card draw more power, even if I disable the onboard one?

I'd have to see the effect played out to know how similar it is to mine. In my case I'd see lots of overruns, a sign the driver isn't able to get the data out as fast as it's arriving, then the IRQ disabled and link lost/up messages. It didn't seem to affect other systems on the same switch, though I was only really watching errors on the sender and receiver.

 

If it helps, mine wouldn't even handle 100Mb gracefully. The problem took longer to show but the results were the same. If you wanted to be more certain of a card fixing your problem you could try a dirt-cheap 10/100 card. If it works it'll be better than the Realtek Gb and maybe we'll get lucky with release 5.0.

 

Yes, the card is bound to use slightly more power than the built-in but it shouldn't be noticeable without test instruments.

OK, well if I'm going to buy a new card, I'm going to go all out and get a gigabit, since I want to have those speeds if possible.

 

I'm going to pick up that Trendnet switch while I'm at it...

 

Do you guys use Jumbo frames? What is the advantage to this? Transmitting larger blocks of data over the network?

Not here, not at the moment.

 

Jumbo frames mostly help where there's lots of latency which makes many small transactions expensive. For a local network that's usually not the case. They can be difficult to make work right depending on the combination of network equipment involved, and changing equipment can mess up your carefully tuned settings.

 

Do jumbo frames last, after everything else is stable.

Intel GT Pro and Trendnet 8 port switch ordered!

 

Being the cheap bastard I am, went with free shipping, which means they probably won't be here until next week.

Interesting note - the Realtek 8111E drive on my mobo that was having the problems, showing Half Duplex support for full GB, check on a Windows 7 system that uses exact same setup (both boards/cpu/ram where purchased at same time). Windows 7 only show support for GB at full duplex - no half duplex option. So, I wonder if it isn't so much the chipset itself, but, the driver included in this version? Might 5b6a have a better drivers? Might the driver support for Realtek just not be that great?

 

Since gone with the PCI-E NIC, not a single issue...

 

Shawn

 

Yeah, its been mentioned the drive could be at fault.  I'm on 5.0b6a and having issues related to intensive file transfers.  Haven't really tried anything else, so I'm saying fuck it, and the NIC should be here soon lol.

Hi,

 

I found this thread trough the wiki, so sorry for bringing it back up.

Anyways, I have the asus e35m1 motherboard ( the pro version).

 

How can I test whether or not my server is having issues with 5b10?

I usually don't do anything intense with my server, but I'd like to still make sure that the server is working fine.

It's pretty easy. Get a parity check started, log in at the console (not via network), tail -f /var/log/syslog, and then copy large files to the server. The two board I've tried would throw errors in a few seconds at 100Mb, crashing eventually. Everything happens faster at Gb speeds.

 

The problems will show eventually even without the parity check, it just takes longer.

 

They've also bad compatibility issues with PCIe SATA cards.

 

They've released a few newer BIOS versions. I've used the first (April) 1002 which didn't help. I haven't yet tried the July 1002 version so can't say. I'm currently running a 2nd-hand copy of version 0091 which I assume is still in development. It's been stable in my limited tests but the speeds aren't great. I haven't tried any of my PCIe cards with it yet.