DSM-G600, DNS-3xx and NSA-220 Hack Forum

dickeywang wrote:

OK, here is the story. I have a Windows desktop and a DNS-323 connected to a D-Link DSG-2205 gigabit switch that supports jumbo frame up to 9600. I changed the frame size from 1500 to 4500 in the desktop's control panel(there is not much choice for the gigabit NIC that comes with the Nforce4 motherboard), and then I changed the DNS' MTU uses the following command:

ifconfig egiga0 mtu 4500

When copying a 606MB file from the DNS to the windows machine, I can see the transfer uses 20% of the gigabit NIC's bandwidth (200Mbps?), while it was something around 15% when MTU=1500. And the time takes to finish the copying is 25sec vs 33sec.
The write speed increases too but not by that much. write the same file from the PC to the DNS uses 15% of the bandwidth(compare to 13% with MTU=1500), and it takes 34sec to finish the transfer compare to 41sec before.

Haven't try any higher MTU values yet, and since the switch is connected to my wireless router(linksys wrt54gs), I don't know if it will have any problem when wireless kicks in.



PS: the D-Link doesn't say that the DNS supports jumbo frame, so I am not sure if it will void the warranty.

Man from the time you put fun_plug on it, your warranty was gone - but - here's the fun_part pull the disks before you send it back and the evidence is gone.

On a more serious note - that jumbo frame is good news, I'll be taking a look at it myself tomorrow.

I recall seeing that setting too. Must have been in the 1.02 firmware because I can't find it in the 1.03 anymore.

I see everyone using an MTU of 4500.  Each of my devices support jumbo frames up to at least 9000. 

Does larger MTU equal faster throughput or should I go with 4500?

Macbook: 9000 (max)
DGS-2208 8 port GigE Switch: 9600 (max)
DNS-323: 9000 (9004 max)

I set them all to 9000, the DNS-323 shows up as 9004 when using the web interface to set the MTU if you then check it from SSH Term using ifconfig.

I don't know if 8 bytes will make a difference.

(edit)

I just rebooted all devices and an getting about 22-26 MB/s xfer rate.  Its a lot better then I was getting before :-)

Thanks
BC

Last edited by BobbyCarbn (2008-10-04 20:18:20)

fordem wrote:

You see everyone using an MTU of 4500??  Where?? How??

I've never used at MTU of 4500, in fact, if I recall correctly 4500 is not an option on any of my equipment.  I suggest you test throughput at the various options and see which gives you the best performance, it might not be the same setting that gives me the best performance.

Take a look at the first post in this thread.  I was just wondering if 4500 was the setting (some) people had seen the best results with.  And, with my macbook and some other hardware I have running at home, I can set the MTU to anything, so I guess we don't have any of the same equipment.

Isn't the whole purpose of a forum or any community for that matter, to save some time now and then by finding out what settings or set up others found worked good for them? 

I agree it is best to test everything out in your environment to see what works best.  But the reality is, some us don't always have the luxury to run benchmarks and tests all day long. 

Although, I do wish I had that kind of time to "get my geek on" on a daily basis.

Last edited by BobbyCarbn (2008-10-06 07:44:12)

Just so you understand the post...

I've pretty much followed up every thread on the use of jumbo frame, so I think I have a reasonably good idea of what they contain, and I don't have any recollection of any one MTU setting being mentioned more frequently than others.

If you check you'll find that you cannot even set the DNS-323 to an MTU of 4500, it starts at 3000 and increments by 1000 until it gets to 9000 - I have no idea what the other forum members use for ethernet adapters in their PCs, I have Intel nics and they don't offer 4500 as an option (4088, 9016, 16128).  I don't think 4500 is a common option.

On testing - based on my experience - you can run transfer speed tests using all of the options the DNS-323 offers in under an hour, not that you would either want or need to - you should have an idea, based on the options available on the other hardware of what you might be likely to use - given the options my Intels offer - the only settings that make sense testing would be 4000/5000 & 8000/9000 - shouldn't take you more than 45 mins.

Sometime you just have to put aside an hour or so to figure things out - it took me three tries to actually get jumbo frame to work, and some of what frustrated my efforts was "mis-information".

The nature of the web lends itself to "mis-information", anyone can post anything, some people deliberately set out to mis-inform and others do it through repeating mis-information that they believe to be correct, in an effort to assist. 

As an example - if you research jumbo frame, you'll come across words to the effect that you must use the same frame size on your entire network - or - that the entire network must support jumbo frame for it to work.  I believe this to be a corruption (or misinterpretation) of "the entire network path must support jumbo frame for it to work" - which really states that the network path between two devices must support jumbo frame for jumbo frame to work between those two devices.  For clarity....

Router<-------------------,
DNS-323 <=======>switch<========>PC1
PC2<----------------------'    '------------------>PC3

In the network above you can use jumbo frame between the DNS-323 and PC1 provided the switch in between (the network path) supports jumbo frame, but the rest of the network need not support it - I actually debated this with a moderator on the D-Link forum who insisted that the result would be fragmentation at the router slowing internet traffic down.

At the time I did not have a switch that supported jumbo frame, so I went out and bought one and created a network very similar to that above and ran my own tests - end result is I know that jumbo frame can be used on a network where only part of the network supports jumbo frame.