Jumbo Frames

[trueimage]

Hey,

 

I'm trying to optimize my home network, and get Jumbo Frames working to get the most out of my network.

 

I'm quoting some info from this page: http://darkwing.uoregon.edu/~joe/jumbo-clean-gear.html

 

My connection goes like this:

 

Cable Modem -> Bufalo DD-WRT router -> Netgear GS108 and Vonage router -> HTPC (Vista 64-bit), unRAID NAS (Linux), Main PC (Vista 64-bit), xbox 360, Wii with usb ethernet adapter.

 

I have a Netgear GS108 switch which is MTU 9216

In the unRAID NAS and the HTPC and Main PC, I have Intel Pro 1000 PT (PCI-X X1) which according to the above link are MTU 16110.

 

My understanding is that the MTU of every device has to be the same. I used the following to set the MTU on the NAS: ifconfig eth0 mtu 9216 up

 

In windows, however, with the newest drivers, the only Choices I have are Disabled, 4088, 9014

 

I'm wondering how I can set this up to work together. It says in the notes for the windows driver:

 

Enable Jumbo Packets only if devices across the network support them and are configured to use the same frame size. When setting up Jumbo Packets on other network devices, note that different network devices calculate Jumbo Packet sizes differently. Some devices include the header information in the frame size while others do not. Intel adapters do not include header information in the frame size.

 

I can also connect the xbox 360 / Wii directly to the router if necessary.

 

Any help in this case would be great, I'm in a little over my head I think.

 

[trueimage]

I'm getting SLOW speeds using Total Commander to copy over files.

 

Did a 4GB file and it took about 8-10 mins, I forget the speeds but they were all over the place

 

from 5,000 to 14,000 kbytes/sec

 

I did a local disc to disc transfer on my htpc and was getting about 60,000 - 70,000

 

 

So, I adjusted the MTU:

root@Tower:~# ifconfig eth0 mtu 9014 up
root@Tower:~# ifconfig eth0
eth0      Link encap:Ethernet  HWaddr 00:1B:21:14:E4:82
          inet addr:192.168.1.134  Bcast:192.168.1.255  Mask:255.255.255.0
          UP BROADCAST NOTRAILERS RUNNING MULTICAST  MTU:9014  Metric:1
          RX packets:4397231 errors:3439 dropped:0 overruns:0 frame:1880
          TX packets:1419027 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000
          RX bytes:1819845027 (1.6 GiB)  TX bytes:348755998 (332.5 MiB)

 

I ran these commands right after eachother so the errors were on MTU 9216 not 9014.

 

The point is, it shouldn't be this slow!

 

I'm tempted to totally re-build and of course I'm going to implement a cache drive but still... I want to be sure my network is working as well as possible.

 

Any help would be appreciated.

 

[jimwhite]

Hey trueimage... do you really look like that?  I know I don't !!! 

 

[RobJ]

Jumbo frames are a great idea, providing better performance with decreased network traffic and lowered CPU usage, but they just don't seem ready yet.  There doesn't appear to be the interest level and effort to achieve the standardization and interoperability to make it work.  Your story seems typical of others: equipment is purchased, and time is spent, resulting in little to no gain, and a lot of frustration.  I believe Tom tried almost a year ago, with little luck.  I suspect he would like to add built in support for jumbo frames, when possible.

 

Your best chance is with equipment (cards and routers/switches) of the same brand and generation, configured identically.  But until I see confirmation of at least a 20% improvement with a given set of equipment, I don't think the effort is worth it yet.  Each attempt, like yours, is useful, to push the story along, and I expect it to work, some day.

 

I too think using values of 9000 and 9014 give you the best chance of any success.  Does the GS108 have a 9014 choice?

 

[trueimage]

So should I just go back to 512 or whatever the default is?

 

The switch is an unmanaged one, it is just preset. That number is from the website I linked in the first post, I don't know if there is a way to set it or anything.

 

I am confused about the padding etc, I need a windows tool to show me TRUE mtu value I guess, and start from there.

[RobJ]

I'm getting SLOW speeds using Total Commander to copy over files.

 

Did a 4GB file and it took about 8-10 mins, I forget the speeds but they were all over the place

 

from 5,000 to 14,000 kbytes/sec

 

That looks typical of a large write to a parity protected unRAID data drive.  That's the 'parity penalty' that the Cache drive solves.

 

Total Commander uses a running average of speeds, over a recent window of the transfer speeds.  In reality, a transfer to a parity protected drive is somewhat faster in spurts, then stops dead for a little, before resuming again.

 

Because many of my batches of large gigabyte transfers occur on my old and slow machine, and its gigabit card uses a LOT of CPU, I like to use the Total Commander copy queuing with speed throttling.  After selecting one or more files for copying, and clicking the Queue button, I immediately click the 'Speed Limit' check box, then adjust the speed.  Since the true overall average write speed seems to be just under 10MB/s, I tend to set it to 8000KB/s or 9000KB/s, for a smooth transfer, in the background.  You can adjust it dynamically, and add more copy jobs also, to the copy queue.  Your results may vary.  If I want more CPU available locally, I'll adjust it down to 5000KB/s, or even 3000KB/s.

 

So should I just go back to 512 or whatever the default is?

 

I think I would try setting both ends to an equal value of whatever is available, 9014 probably, and if I couldn't see any improvement, let them revert to their defaults.