Hello,

 

I've got a gigabit network setup (simplified diagram : [unraid] <--> [switch] <---> [Windows 7])

 

I noticed that transferring files from the windows 7 box to the unraid box is slow (~150-200 mbps), while transferring files from unraid to windows are ~400-500 mbps...After some investigation, I noticed that on the unraid box I'm getting lots of rx overruns.

 

So because of this, I think the rx overruns are the 'root cause'...

 

I've tried:

1. Changing cables

2. Tweaking rx_mem variables

3. Changing the size of the mtu

 

But it doesn't seem to have helped....I thought the motherboard was the issue (since it was an integrated NIC), but if it was, would I be getting such high tx rates?  Would adding a cache drive help? Is there some way to drastically increase the rx buffer so that it doesn't overrun? I saw some old linux posts talk about IRQ conflicts, but I don't notice any; how can I tell?

 

 

Here's some information about my setup:

 

* Unraid 5.02beta, with 2 drives and 1 parity drive.

* I forget the make/model of the motherboard, but it's around 3 year old p4 running 3Ghz, with 3gig of ram

* Here's info on the ethernet card:

 

root@Tower:~# mii-tool -v eth0

eth0: negotiated 1000baseT-HD flow-control, link ok

  product info: Yukon 88E1011 rev 5

  basic mode:  autonegotiation enabled

  basic status: autonegotiation complete, link ok

  capabilities: 1000baseT-HD 1000baseT-FD 100baseTx-FD 100baseTx-HD 10baseT-FD 10baseT-HD

  advertising:  1000baseT-HD 1000baseT-FD 100baseTx-FD 100baseTx-HD 10baseT-FD 10baseT-HD flow-control

  link partner: 1000baseT-HD 1000baseT-FD 100baseTx-FD 100baseTx-HD 10baseT-FD 10baseT-HD

root@Tower:~#

root@Tower:~# ifconfig eth0

eth0      Link encap:Ethernet  HWaddr 00:13:d4:05:3f:42

          inet addr:192.168.2.4  Bcast:192.168.2.255  Mask:255.255.255.0

          UP BROADCAST RUNNING MULTICAST  MTU:2000  Metric:1

          RX packets:1898887 errors:0 dropped:0 overruns:25873 frame:0

          TX packets:3054784 errors:0 dropped:0 overruns:0 carrier:0

          collisions:0 txqueuelen:3000

          RX bytes:2330213128 (2.1 GiB)  TX bytes:3185654876 (2.9 GiB)

          Interrupt:22

 

 

Thanks,

 

rwc

If you can get the make and model of the motherboard that would be a huge help, along with a syslog.  See the  troubleshooting link in my signature on how to capture a syslog.

 

The thing I see right off the bat is that your board is using a Yukon NIC card.  These and Atheros tend to be the pain in the arse NIC's for unRAID.  My suggestion is to email Lime-tech to see if a new driver will be included with 5.0beat3 (due for release any day now), and if that does not solve the problem find a PCI NIC card using an INTEL chip.

Hi, thanks for the help!

 

I've attached my syslog....I'll email tech support as well....

 

Later tonight I should be able to get the make/model of the motherboard...

syslog-01-25-2011.txt

So because of this, I think the rx overruns are the 'root cause'...

 

I think you are right, but there may be additional network-related problems too.  Your syslog shows the network went down 3 times in the relatively short period covered by the syslog.  The first 2 times were only for 3 to 4 seconds, but the 3rd was down for about 35 seconds, quite a long time, unless you are aware of something that would explain it, like rebooting the router or reconnecting network cables.  It's possible the driver was alarmed by all the overruns, and was trying to reset the network chipset for better operation, those 3 times.

 

I'm not an expert here, but my sense of overruns is that they should no longer be a problem with modern hardware.  And you have an Intel ICH4 based system, with a 3.2GHz Pentium 4, perhaps a little old but more than capable enough.  I checked the IRQ usage, and nothing else is sharing IRQ 22.

 

I'd recommend installing a gigabit card, preferably PCIe based, but PCI if you have to, and disable the onboard networking.  I can't guarantee it, but that should work correctly.

I'm pretty sure those times were me trying various settings and/or changing cables, so I'm not too surprised there

 

Maybe I'll get lucky and hear some good news from support I'll let you know...thanks for your help!

 

rwc

 

As it is, there should be a lot of new(er) drivers included in the 5.0 beta 3 simply due to rebasing unRAID off a recent version of Slackware, perhaps 13.1 or maybe even a 13.2/Current version.

 

 

By the way, I forgot to mention that your syslog indicated that the Reiser subsystem had fixed some minor issues in the Reiser file system on Disk 1.  I would run reiserfsck on Disk 1 to make sure everything is now fine -  see Check Disk File systems.  I don't expect it to find any problems, but just in case ...

Try 4.7 build, it has some Win 7 fixes with too many open files I believe.

 

rx overruns means the server was unable to handle all the incoming packets. First off set MTU back to 1500 unless your switch can handle packets over 1500 (jumbo frames), which I doubt, even if it can turn them off for now.

 

Make sure your Win7 machine doesn have jumbo frames enabled. MTU set to 1500 (std). 

 

 

2000 seemed to work well (via trial and error)..the switch and the computers all support jumbo frames.  Letting the frames get too big (EX: 9000) definitely kills the unraid box though...

I think I know the answer, but is there any way to "install" an updated network driver into the unraid kernel? Or does it require a whole recompile/reconfigure?

 

Also--not that I'm holding anybody to any dates--but when they say 5.03beta is due "any day"; should I wait (ie: most likely in the next week or 2?) or should I just bite the bullet and get a new NIC?

I think I know the answer, but is there any way to "install" an updated network driver into the unraid kernel? Or does it require a whole recompile/reconfigure?

A whole recompile is needed

 

Also--not that I'm holding anybody to any dates--but when they say 5.03beta is due "any day"; should I wait (ie: most likely in the next week or 2?) or should I just bite the bullet and get a new NIC?

5.0b3 was just release, BUT there appears to be some significant changes that are posing problems/huddles for those running 5.0b2.

 

I would not run 5.0b3 on a production server as of yet.  There were a lot of underlying changes, you can try to install the new version and see how your onboard NIC performs but I would not run 5.0b3 longterm unless you are willing to deal with some of the addons/etc not working.