Jumbo Frames. Do I need them?

[msjpt]

Hi all,

I have a small office network with no server. Just 5 desktop Windows 7 computers and 2 network printers, all setup as a Windows Homegroup network. Regarding network hardware I have the following:

ISP Modem > Gigabit Router Draytek 2130 > Gigabit Switch TP-Link TL-SG1005D

I also have a Western Digital My Book Live connected using Gigagit LAN to my router. This is where all the desktop pc’s save information. It’s used as our company shared hard drive.

Regarding Jumbo Frames, I don’t know exactly what they are and if they will bring any advantage on my network. I also don’t know if my Draytek router and TP-link switch support Jumbo Frames.

So my request to anyone who is reading this, is to explain in a simple way what are Jumbo Frames and if on my current network setup enable them will bring me any advantages and/or benefits?

Also please try to tell me if my Draytek Router and TP-Link Switch both support Jumbo frames.

Thanks

 

[L&LD]

http://www.smallnetbuilder.com/wireless/wireless-reviews/31132-draytek-vigor-2130n-reviewed

http://sotbadness.wordpress.com/2009/05/17/tp-link-tl-sg1005d-gigabit-switch-review/


Seems like Jumbo Frames are supported.

The speed increase is what advantage you'll see (if it works with the rest of your devices (make sure to enable Jumbo Frames on each client PC).


With a network that supports Jumbo Frames; yes you need it for best performance.

 

[Dennis Wood]

On Gbe you may see little difference using jumbo frames. I believe this is due to CPU and sata6 disk IO generally not limiting speeds on gigabit networks. During 10Gbe Ethernet testing however, jumbo frames do help in the windows environment, likely due to other limiting factors related to latency. For the first time in my networking experience, at least at 10Gbe, jumbo frames do make a difference. Using jumbo frames reduces CPU overhead and reduces data transferred as well as tcp "packaging expense". Going from 1500K Ethernet packets to say 9000K packets basically takes 6 packets of data and replaces it with one larger packet. In slower machines (like older NAS units) it may be worth testing.

Doug Reid did an excellent summary here at SNB: http://www.smallnetbuilder.com/lanw...1-need-to-know-jumbo-frames-in-small-networks

 

[msjpt]

Hi all,

First i want to thank you all for your advices.

So my Draytek router and TP-Link switch both support jumbo frames. All my desktop PC's are modern computers with Windows 7 and Gigabit LAN, so i supose all them also support Jumbo Frames.

But on my network i also have 2 network printers with 100Mbps LAN connected to the TP-Link Switch and a WD My Book Live connected using Gigabit LAN directly to my Draytek router.

So from what i have been reading in order for Jumbo Frames to work properly, all the devices on the network must support them. So if this is true, and taking as example my own network where all the router, switch and computers support Jumbo Frames but i guess my Printers and also WD My Book Live does not.

So will the printers and external disk, will have a negative impact if i enable Jumbro Frames?

Because of the network printers and external disk, i should not enable Jumbo Frames on my network?

Any advice?

Thanks

 

[L&LD]

If I'm not mistaken, the WD MyBook does support Jumbo frames.

As for the printers not supporting it; as long as they continue working with Jumbo Frames enabled with the rest of your devices, they certainly can't get slower (than the LAN).


All you can do is try. Won't break anything and in my experience; it can improve throughput by 40% or more between SSD based laptops.

 

[msjpt]

If I'm not mistaken, the WD MyBook does support Jumbo frames.

As for the printers not supporting it; as long as they continue working with Jumbo Frames enabled with the rest of your devices, they certainly can't get slower (than the LAN).


All you can do is try. Won't break anything and in my experience; it can improve throughput by 40% or more between SSD based laptops.

Hi L&LD,

Once again thanks for your advice. So just to resume, the fact of having network printers connected to the same switch will not have any negative impact if i enable Jumbo Frames, because all my other network devices (router, switch, computers and my WD My Book Live) are compatible with Jumbro Frames, correct?

Thanks

 

[L&LD]

Yes, that is how it should work. Barring any weird interactions between the router, switches and printers.

Always test methodically and then forge ahead.

 

[azazel1024]

You might see an increase because the WD Mybook is likely pretty CPU bound already.

Between my server and my desktop, I see little difference enabling jumbo frames. Same with on my laptop. Desktop will do 117MB/sec regardless of frame size and between regular MTU and 9k I see a 1% drop in CPU usage, no speed difference. The laptop is the same (though it only chugs along at 114MB/sec regardless, Realtek NIC where as the server and desktop are using Intel NICs). My T100 is a wonky situation as the eMMC in it is very write limited and read can't quite saturate a link, that and the USB3 adapter is just wonky in general. Write speed seems to get a slight bump from 28MB/sec to 30MB/sec going from regular to 2k, but 4k just locks up CIFS. Can't browse, can't transfer files. Regular TCP and UDP traffic works fine, it just seems to be CIFS that gets borked. Read speeds are roughly the same 100-105MB/sec regular or 2k.

On my desktop with paired NICs and SMB multichannel I DO see a slight increase in throughput. Near the outer bits of my RAID0 array it chugs at 230MB/sec regular MTU, but with 9k Jumbos it'll go up to 235MB/sec or so. So a slight difference there, but CPU usage drops about 2% in the case of 9k jumbo and paired NICs.

 

[Dennis Wood]

What is surprising, are the results below. One plot is with both target/host 10G Intel x540 NICs using default server settings, the other is with all performance options tweaked and jumbo frames enabled. The target is a windows 8.1 box with a rocketraid 2720 and drive array of 6 x 4TB Deskstar 7200 rpm drives in raid zero. These results reflect pretty much exactly what I'm seeing in windows large file copy/paste tests. The only change in between tests is the NIC settings. Pretty surprising results if you're used to little/no differences in driver/jumbo frame tweaks on gigabit...

The ATTO benchmark btw is pointing to Q: which is a network mapped drive. Using the subst command in windows, ATTO will see that drive as local. Very handy for testing over 10G. Both workstations are running windows 8.1 pro and therefore SMB3 which is key to these results. Windows 7 just won't do these numbers with SMB2.

Default settings:

Jumbo frames and performance tweaks