In reply to a post by billford:

Playing with the values might be worth a try, but I'd leave the router at 1454 (because of the max before fragmentation of 1426 that you've found)

Thanks - but I thought the maximum size before fragmentation of 1426 was because the router MTU was set to 1454, borne out by the fact that now I've changed it to 1492 as an experiment, the max size before frag is 1464.

In reply to a post by kasg:

Thanks - but I thought the maximum size before fragmentation of 1426 was because the router MTU was set to 1454, borne out by the fact that now I've changed it to 1492 as an experiment, the max size before frag is 1464.

Ah, right, I didn't realise that.

It might be worth worth trying 1500 on the router, 1492 on the PC to see if that works, but any improvement in performance is going to be so slight that it'll likely be swamped by normal internet variability and I'd be inclined to leave it where it is

As a (slightly) side issue- in general it's better to have the highest value of MTU that will work, but some sites don't like "big" ones- I think Microsoft may have had this oddity at one time. So it can be worth sacrificing a (tiny) bit of performance in case you happen to come across another one.

I found this out big time when I was using an Airport Extreme's built in 6to4 tunnel for IPv6- that adds a lot of overhead, and quite a few sites wouldn't respond to a tunneled connection if the frames were fragmented... eventually I got fed up and just set the computer's MTU to the minimum allowed (~1328 iirc)

Yes, I know all that.

I was trying to understand your statement "PPPoE uses 8 bytes more than PPPoA (added by the router)" and relating this to the relative sizes of MTU (the 'M' does stand for "Max") on router and PC.

It's as if you are saying that a packet comes over the ADSL cable (PPPoA) from exchange and the router adds 8 bytes in order to pass it over the Ethernet cable (PPPoE) to PC. Thus the packet is bigger at the PC than on the ADSL. Therefore the MTU at the PC needs to be bigger than the MTU at the WAN i/f of router.

Then your statement "the MTU on the PC should be set to 8 less than on the router" doesn't follow and is a contradiction.

"there is only a single MTU per adapter. "

Sort of.

The user specified per-adapter one is the maximum maximum, nothing on that adapter can or will exceed this.

But in the bigger picture there is Path MTU discovery, where traffic to a given destination (via a given route) may have another smaller MTU, which may supercede (for that path) the per-adapter value.

Good game, innit.

In reply to a post by XRaySpeX:

Yes, I know all that.

I was trying to understand your statement "PPPoE uses 8 bytes more than PPPoA (added by the router)" and relating this to the relative sizes of MTU (the 'M' does stand for "Max") on router and PC.

Yes, and the 'T' stands for 'Transmission'... so you were ignoring the direction of the traffic.

The PC assembles the frame and then sends it to the router, which adds 8 bytes before sending it off to the modem for conversion to whatever the ATM requires. (Whether it adds these bytes if it's LAN traffic I don't know).

It's the job of the PC and router to ensure that the frames it creates meet Ethernet requirements, for incoming traffic the size of the frame is what it is and the router and PC just have to make the best of it. If it's greater than that allowed by the Ethernet specs then I assume it will get dropped.

In reply to a post by billford:

sends it to the router, which adds 8 bytes before sending it off to the modem for conversion to whatever the ATM requires.

Where the 8 bytes are removed cuz "PPPoA uses 8 bytes less than PPPoE"?

Otherwise PPPoA is using 8 more bytes than came from PC over Ethernet (PPPoE)?

Perhaps please clarify where the extents of these Point-to-Point Protocols apply?

In reply to a post by XRaySpeX:

Perhaps please clarify where the extents of these Point-to-Point Protocols apply?

As far as I'm aware- from router to modem to modem to router, ie whilst it's going over the internet. Once the data is within an Ethernet network they don't apply.

(edit- I think, not sure, that they would also apply between two LANs, each of which required its own authentication, IP assignment, traffic monitoring etc etc. ie it's used over an inter-net connection, not only "the internet")

But all the modem does is to convert signals between one form and another (Ethernet and ATM), it doesn't do any processing on the data carried by those signals so it has to be correctly assembled first. By the end which is sending (transmitting) them.

Edited by billford (Mon 16-Jul-12 14:12:08)

I said earlier that for incoming traffic the router has to make the best of what it gets- there's also an MRU (Maximum Receive Unit) set in a router, but not normally available to the user (afaik).

It should be 1500 to handle any legal Ethernet frame, but it can be set to a lower value. If a a frame arrives that is larger than the MRU then the routers won't be able to communicate, I've had this happen in the past on occasional sites.

In reply to a post by Anonymous:

"there is only a single MTU per adapter. "

Sort of.

The user specified per-adapter one is the maximum maximum, nothing on that adapter can or will exceed this.

But in the bigger picture there is Path MTU discovery, where traffic to a given destination (via a given route) may have another smaller MTU, which may supercede (for that path) the per-adapter value.

Good game, innit.

Yep agreed - not all OSes support Path MTU discovery (PMTU) correctly - and whats worse many *security* products break it as well (they tend to block the ICMP status returns or block the outbound ICMP requests).

As others have said some routers always assume PPPoE is in use and have a limited MTU to 1492 which in the UK with PPPoA on normal ADSL at 1500 shouldn't be needed - but for FTTC on WBC networks using PPPoE then I guess the 1492 makes sense.

Sky FTTC doesn't use PPPoE as they use IPoA (similar to BE) with their proprietory authentication and TalkTalk no idea.

Getting back to your Q ....

When you observe your maximum packet size reducing does the lower size +1 need fragmentation when you do 'ping -f -l nnnn bbc.co.uk'?

You did ping w/out '-f' earlier and that just timed out. I can't see that as a lowering of MTU cuz it should still be able to fragment.