Setting MTU via DHCP

johnpoz

I can tell you that for all the corps had to support over the years.. Nobody is doing this on their local lan that client machines are connected to.. There have been a few companies that had jumbo setup on specific segments (waste of time if you ask me)..

This just makes no sense on your typical lan, especially a home network.. If want to play with it in your "lab" have at it.. Its not going to get you anything other then headaches when something doesn't work ;)

But yeah if windows shop was going to do it - they would prob set via group policy sure.

Here is a simple test... Set your storage nas/server with the larger mtu you want... Now set your machine your going to move files back and forth in your test... Move some files.. Did you speed increase by X percent? Did your cpu usage lower by Y percent.. Do the % gained in either of those two measurements make it worth the hassle? My bet is NO! ;)

JKnott

@kiokoman

It might not be common today, but years ago, when DHCP was starting out, different MTUs were common. I recall 576 used to be frequently used and, as I mentioned, we used 4000 at IBM, but it could have been configured for other. Still, a DHCP client that doesn't configure correctly for the network it's connected to is defective.

BTW, a Google search shows many others asking about why Windows doesn't support option 26. I guess this is more of Microsoft's habit of ignoring specs and doing their own thing.

In an earlier post I mentioned jumbo frames are used in the Internet2, which is used by a lot of researchers. This means they can't plug a Windows notebook directly into the network and then into other networks, without reconfiguration.

JKnott

@johnpoz said in Setting MTU via DHCP:

I can tell you that for all the corps had to support over the years.. Nobody is doing this on their local lan that client machines are connected to..

I'll have to ask my cousin what he does. He's a nuclear physicist, who works with neutrinos. He spends a lot of time on supercomputers and runs Red Hat Linux on his notebook computer. He's currently at the Sudbury Neutrino Observatory, but had previously been at Fermilab and a couple of other places.

kiokoman

I wasn't trying to find an excuse for the crap behavior of windows
for example now that i have my stratum 1 running in a raspberry i would love to see option 42 but even that is not possible

johnpoz

Yeah sure there are questions about why windows doesn't ask for mtu in dhcp.. Not really sure what that has to do with the just plain FACT that its not worth the effort to change this on your local home network... If your a geek/nerd and into this sort of thing and you want to play with it - have at it on your lab network.

I went through my lets play with jumbo years ago when gig was somewhat new to the market, jumbo was all the rage - lower the cpu needed to move stuff, etc. etc.. Yeah back then a lot of the stuff was having to be done by the cpu, and not offloaded onto the nic, etc.

CPUs were far less powerful back then... Take it or leave it - this is my personal and professional opinion, with 30 some years in the biz, and plenty of geeking and nerding about network both at work, in my home, labbing this and that, etc.. etc..

Playing around with jumbo isn't worth the effort.. its not going to get you anything but headaches.. There just is no point to changing from the default.. You want to shave off your cpu from doing a couple of cycles?? So it can just do more work when wanting to talk to anything else on your network that is not capable of changing the mtu to jumbo.

Internet 3 going to use super jumbo frames ;) Doesn't mean anything when your talking about some local work/home network..

JKnott

@johnpoz said in Setting MTU via DHCP:

But yeah if windows shop was going to do it - they would prob set via group policy sure.

Does Active Directory support assigning MTU size?

As I mentioned, there has been some discussion about moving the Internet to jumbo frames, as the conditions that drove the smaller MTUs no longer exist. Also, users are moving a lot more data than we were 20 - 30 years ago.

From a network/ISP point of view, larger MTUs mean a lot less work for routers, which is a major consideration for many companies.

johnpoz

@JKnott said in Setting MTU via DHCP:

Does Active Directory support assigning MTU size?

It does via making changes to the registry.. Which yeah pretty simple to roll out any sort of reg entry you want.

larger MTUs mean a lot less work for routers

If their routers are over worked moving their data around their network, their routers are under sized ;)

JKnott

@johnpoz said in Setting MTU via DHCP:

Not really sure what that has to do with the just plain FACT that its not worth the effort to change this on your local home network...

Perhaps not on a home network, but what about places like Google, where they have to move a huge amount of data? Each packet requires the router to handle it and that's done with something called interrupts, where the NIC taps the CPU on the shoulder and says handle this. The CPU then has to stop what it's doing, save the current state, handle the interrupt and then go back to what it was doing. With 1500 MTU, that happens 6x more often than with 9000. Then there's also the reduced bandwidth demand by having 1 header in place of 6, though that's a minor consideration. The Internet is getting faster (some carriers and ISPs are now moving to 100 Gb) and they need some way to reduce the load on routers, Jumbo frames help with that.

JKnott

@johnpoz said in Setting MTU via DHCP:

It does via making changes to the registry.. Which yeah pretty simple to roll out any sort of reg entry you want.

What happens with a notebook, where the user takes it to a another network? Does that registry setting still apply?

BTW, my only experience with AD is resetting passwords.

JKnott

@JKnott said in Setting MTU via DHCP:

As I mentioned, there has been some discussion about moving the Internet to jumbo frames, as the conditions that drove the smaller MTUs no longer exist.

Jumbo Frame Deployment at Internet Exchange Points (IXPs)

Hurricane Electric is a major Internet backbone provider.

kiokoman

yes the registry will not change, at this point you can create a simple .bat script that will change that value based on where you are or buy a second network card to use only with mtu 9000 set it and forget

JKnott

@kiokoman said in Setting MTU via DHCP:

yes the registry will not change, at this point you can create a simple .bat script that will change that value based on where you are or buy a second network card to use only with mtu 9000 set it and forget

And you're back to having manually change something that should be done automagically. As I said, a DHCP client should adjust itself to the network, without any user intervention. This is definitely a "feature" in Windows.

I was at IBM when DHCP started to become popular. I had my own static ip (9.29.146.147), which I used in my office, but I had to reconfigure when I went to another site. Someone at IBM came up with a utility for OS/2, which I could use to select different profiles, according to where I was. However, the switch to DHCP removed the need for that, other than switching between my static IP and DHCP. Funny thing, my current notebook computer, running Linux, still has that ability, where I can run a variety of profiles. As someone who frequently connects to customer networks and often needs a static configuration, that is quite handy, but it's not available in Windows.

kiokoman

that was an expired draft from back 2011
it's interesting what is written in the attached pdf https://ripe63.ripe.net/presentations/129-Jumbo_Frames_RIPE63_Nov2011.pdf

Pros and Cons: summary
Cons
1.No standard/agreement for size of Jumbo Frames
2.Increase transmission time, packet delay, jitter, etc.
3.Require bigger buffers on equipment
4.PMTUD doesn’t work
5.Low traffic with the current maximum size

Pros
1.Less CPU load
2.Less network packet overhead
3.Better TCP performance

30 4. Pros and Cons: applications

Cons
1.Inter-process communication (IPC)
2.Protocols using small packets (DNS, VoIP, etc)
3.Interoperability (no standards, broken PMTUD)

Pros
1.Data transfer (Backups/Clusters/NFS/NNTP)
2.VPNs with payload 1500 bytes
3.SAN (FCoE/iSCSI)

31 4. Conclusion Personal
1.Nature of Internet traffic: small packets
2.All talks about Jumbo Frames are similar to IPv6 talks: started in 90x but IPv4 addresses are over and Ethernet with 1500 bytes still works fine

The performance advantages are small.
The implementation headache can be significant.

In my opinion, Jumbo Frames aren't worth the effort. Equipment vendors have really optimized the heck out of 1500MTU packet delivery.
to gain only maybe 2% in real usage
http://www.boche.net/blog/2011/01/24/jumbo-frames-comparison-testing-with-ip-storage-and-vmotion/
This comparison is from 2011,
With the cpu power we have now i bet is less than 2%
too many cons vs pros

johnpoz

@kiokoman said in Setting MTU via DHCP:

In my opinion, Jumbo Frames aren't worth the effort.

Exactly!! They might have niche use cases.. But in your typical network - especially in a home with all kinds of different devices wanting to talk to each other.. Your media player wanting to pull your movies off your nas going to support them? That wireless roku stick you use to watch your plex or stream your music library, etc.

Your pc printing to your "wireless" connected printer - sure great use case for jumbo there.. Make as BIG as possible ;) You know to shave off cpu time <rolleyes>..

Add up ALL the devices in your network that interact with each other.. So ok you can shave 1/8 of 1% of cpu for the 20 seconds it takes to move a movie from your pc to your nas.. What about all the other devices pulling stuff that nas when the nas wants and things the network mtu is 9000..

If you want to use jumbo on the SAN you use to backup your nas to your other nas - Go for it, might see some advantage there.. But for your overall network - just not worth it..

To windows machine not pulling the mtu from dhcp - really talk about a niche issue.. How many home users boxes running windows even connected to a wired network these days? ;)

Nischi

Thanks everyone for their replies. I didn't mean to stir up any conflicts here regarding use cases. This is purely for my own educational reasons in a home/lab-environment.

JKnott

@johnpoz said in Setting MTU via DHCP:

To windows machine not pulling the mtu from dhcp - really talk about a niche issue.. How many home users boxes running windows even connected to a wired network these days? ;)

Same applies to WiFi, though the max MTU is much less than 9K. As for home users, how many of them connect to Google? Might not Google benefit from jumbo frames? They can't if the user can't use them. Can you imagine what things would be like if we were forced to use 576, as was often the case years ago. Given Internet2 is already running 9000 MTU, there's no reason why the regular Intermet couldn't be switched to it too. I'm not saying there should be a flag day, when everyone switches to 9000, but just stop putting the 1500 limit in. Arbitrary limits should never be forced, as sooner or later they will bite.

This is in some ways the same argument people use for not moving to IPv6. They claim the current situation is OK and no improvement is needed. Well, that's not the way the world works. IPv4 stopped being adequate the day NAT became necessary and given the huge increase in traffic over the years, 1500 MTU has become a performance barrier.

JKnott

@kiokoman said in Setting MTU via DHCP:

that was an expired draft from back 2011
it's interesting what is written in the attached pdf https://ripe63.ripe.net/presentations/129-Jumbo_Frames_RIPE63_Nov2011.pdf

Pros and Cons: summary
Cons
1.No standard/agreement for size of Jumbo Frames

9000 is often used, as in Internet2

2.Increase transmission time, packet delay, jitter, etc.

1500 was the standard back in the days of 10 Mb half duplex Ethernet. Now it's typically 1 or even 10 Gb FD.

3.Require bigger buffers on equipment

Not an issue these days with large cheap memory.

4.PMTUD doesn’t work

It's mandatory on IPv6 and IPv4 is already moving to it. With Linux, PMTUD is used for everything, on Windows TCP now has the do not fragment flag set.

5.Low traffic with the current maximum size

What of all those web pages people read? Are they all under 1500 bytes? Streaming audio/video over TCP? On the other hand VoIP does use small packets.

Pros
1.Less CPU load
2.Less network packet overhead
3.Better TCP performance

30 4. Pros and Cons: applications

Cons
1.Inter-process communication (IPC)
2.Protocols using small packets (DNS, VoIP, etc)
3.Interoperability (no standards, broken PMTUD)

Having 9000 MTU available does not mean it will be used exclusively. As before, packet size varies, according to what's needed and what's supported. So, if you're downloading large files, then large MTU is what you want. If you only need small packets then that's what you use.

Pros
1.Data transfer (Backups/Clusters/NFS/NNTP)
2.VPNs with payload 1500 bytes
3.SAN (FCoE/iSCSI)

31 4. Conclusion Personal
1.Nature of Internet traffic: small packets
2.All talks about Jumbo Frames are similar to IPv6 talks: started in 90x but IPv4 addresses are over and Ethernet with 1500 bytes still works fine

See my previous post, but that's generally a "head in the sand" attitude.

The performance advantages are small.

Tell that to the people who build high performance routers and networks.

The implementation headache can be significant.

Why? 9000 and more MTU is now common on a lot of equipment. If your equipment is not capable, then don't run it. What's beyond the router is irrelevant. Incidentally, I ran a test with 9000 MTU on my network a while ago and the only thing that couldn't handle it was the WiFi AP, due to it's own MTU limitation. I did not test with other appliances, but my computers, unmanaged switch, etc. worked fine. One hard limit for home, small office users would be the WiFi MTU (2304), if the access point is connected directly to the LAN. If it passes through a router first, then it's no longer an issue.

In my opinion, Jumbo Frames aren't worth the effort. Equipment vendors have really optimized the heck out of 1500MTU packet delivery.
to gain only maybe 2% in real usage
http://www.boche.net/blog/2011/01/24/jumbo-frames-comparison-testing-with-ip-storage-and-vmotion/
This comparison is from 2011,
With the cpu power we have now i bet is less than 2%
too many cons vs pros

Funny thing about that PDF. It seems a lot of the objection is "people don't like change". Well that's the case for a lot of things in life. I was just reading a book about time and frequency from the NIST and it mentioned people were even opposed to standard time zones, even before we get to the debate about daylight savings time. People just don't like change, even though that change may make things better for them. I also read in that PDF about the performance calculations, where they considered just the more data per header issue, but not router performance, due to fewer packets.

That PDF also refers to 802.3 frames, when IP runs over DIX II frames, which are similar, but not the same as 802.3. A big difference is there's no maximum payload size or frame length specified for DIX II. 802.3 has that hard 1500 byte payload limit, but with DIX II, the frame ends when the frame ends. There is no field in the frame that says otherwise. The frame sizes listed in that PDF are related to that 802.3 limit, not DIX II. One thing that PDF doesn't mention is that 802.3 requires LLC/SNAP to get the same functionality as the Ethertype frame in DIX II. That counts for a few more bytes. However, these days I doubt you'll find 802.3 being used for much beyond spanning tree. It's certainly not used with IP. And yes, I am well aware that back in the early days memory size & cost were a factor in going with 1500 bytes, but that no longer applies.

BTW, many years ago, I hand wired some Ethernet controllers on Data General Eclipse prototyping boards. Back then chip count really mattered, whereas today Ethernet controller chips are used, instead of the discrete logic as was used on the boards I built.

JKnott

@Nischi said in Setting MTU via DHCP:

I didn't mean to stir up any conflicts here regarding use cases.

We're just getting started!

This is purely for my own educational reasons in a home/lab-environment.

Excellent reason, one that I have often used. You can learn a lot that way.

However, one thing that really gets me are the head in the sand types, who argue against change, because they've always done it that way. That's a real innovation killer.

kiokoman

yes i'm putting my head in the sand but tell me, what do i have to do? i can't set MTU on my phone, i have to manually set MTU on windows, i can't set MTU on my IOT and you can't have mixed MTU device on the same LAN, my ISP does not support 9000
all this trouble to gain maybe 1%
of course i put my head in the sand, to hell with MTU, you have to tell your reasons to all the other ISP out there, to all the manufacturer of phone and iot, modem, router etc etc. when it will be the stardard i will think about it.

@Nischi said in Setting MTU via DHCP:

Thanks everyone for their replies. I didn't mean to stir up any conflicts

we are having fun here, an exchange of constructive thoughts, that's what a forum is for, intellingent people can discuss without hard feelings

This is purely for my own educational reasons in a home/lab-environment.

test test test do something wrong and test again that's the best way to learn, don't trust us do your test and see for yourself

johnpoz

@JKnott said in Setting MTU via DHCP:

head in the sand types, who argue against change,

Dude so your running jumbo on your home network??? Come on - let me know how you set your printer to use jumbo.. Did you hack up some of your own firmware and driver for its nic? Shouldn't it just pick up the mtu from dhcpd - oh wait it doesn't do that either... WTF ;)

This is not head in the sand - this is just common sense, I have no use for this PITA since there is no main stream support for it as of yet... If everything on my network supported jumbo... Then it would be the new standard and we wouldn't having this discussion in the first place.

It's kind of like your attitude towards ipv6.. You have it in your head that its the STANDARD, and required.. No matter how much you want it to be.. It just not here yet.. How many iot devices support it... how many actual isp support it? etc etc. Name 1 major internet resource that I can only get to on ipv6.. You can not..

Its just like this jumbo nonsense... Sorry but it has no use... Maybe when we are all running 40gig on local networks it will be, but currently sorry its just not the case.. How about you get with all the iot makers to get their shit to do jumbos and then maybe we can have a discussion on why not running it ;)

You know what while your talking to them - ask them why they put a freaking 10/100 nic in a 4K tv? What the gig nic would of cost what a $1 more or something? Or also have them support wpa enterprise vs this nonsense psk, so I can run eap-tls on them..

The OP can do what ever he wants... His original question about mtu and dhcp has nothing to do with pfsense at all.. It's a windows issue.. Take it up on their forums if you don't like how they are doing it... But the overall big picture is its pointless anyway.. Nobody is running jumbo in home network - NOBODY!! At least nobody with any common sense, because the gain is not worth the effort and hassle. And in the corp world, if some IT guy had the smart idea to try and run jumbo on the users network.. He wouldn't be working there long ;) Love to be on the cab where he was discussing making that change ;)

I am more than happy to help you test and play with jumbo in your "lab" or for fun - but here is what your going to find out.. Its not fun, nor is provide any actual real world performance gain ;) Its not "fun" because all you run into is headaches on why doesn't this piece of shit support jumbo.. What do you mean this maker thinks jumbo size is this, and this maker thinks its that... You mean there is not actual standard they "have" to follow for the size?? WTF... Yeah this is the future for sure ;) You mean when I pick jumbo on this device and jumbo on that device and they don't match up exactly the flow of data between them is going to be way slower and cause the cpu to do way more work... who would of thunk it ;)

We can all talk about pie in the sky, oh they doing this oh they got this working, etc.. Still waiting on my freaking flying car.. But in the mean time while we wait for that there are actual cars... Just like sure super extra special large jumbo frames might be the thing when we have quantum computers and 100G wireless and 1TB wired connections.. But most of us live in the real world where I just want my TV to play my video off my nas and not stutter and or freeze or cause any issues.. Guess what gig and "standard" 1500 mtu does that.. Because EVERYTHING uses it...