> Ever since I came across Lawrence Baldwin's site, I've been trying to
> work out what the optimal max MTU for a ADSL connection with PPPoA.
> His site discusses PPPoE but he mentions and I think he's right he's
> made a few mistakes and he's given up on it.
> Anyway, I've been looking into it and best as I can make out, for
> VC-mux the optimal highest MTU commonly supported would be 1480. AAL5
> with VC-mux has a 8 byte SAR trailer at the end of the last cell and
> is padded if necessary. With a 1480 byte MTU, 30 of the cells with
> have a full 48 bytes of data, the last cell will have 40 bytes of data
> and the 8 byte trailer.
> With SNAP/LLC, the optimal MTU should be 1472 since it adds a 8 byte > header.
> I think all this is correct assuming I'm right that a 1480 byte IP
> transmission unit means 1480 bytes through the ATM link. I'm still not
> clear if there is any other overhead I'm missing. Lawrence's site
> suggests there is ethernet overhead but I think he's wrong??? If there
> is a 38 byte ethernet overhead that is sent through the ADSL ATM link,
> then the optimal max MTU would be 1490 I think (for VC-mux)?
> For PPPoE, there is another 8 bytes of overhead to consider of
> P.S. This is only relevant if you want to get the maximum out of your
> modem's ADSL link. If, for example, your ISP does not give the maximum
> from your link but does traffic shaping later on, it doesn't really
> matter, in fact a max MTU of 1500 is optimal since this minimises > TCP/UDP overhead.
Sorry it's posted top level - it got rejected as a proper reply.
For pppoa/vcmux in the UK the optimal MTU is 1478 - the pppoa overhead is
- (1478+10) / 48 = 31 whole cells, each which get another 5 bytes ATM header added.
IIRC the RFC says 8 byte trailer and 1 or 2 header. I know what my overhead is from the cell count on my modem and from the fact that I can see less throughput with MTU 1479 compared to 1478.
(linux) ping -s 10 sends one cell -s 11 two. For pppoe users I think sometimes the headers do get sent aswell - which makes any tweak less noticable for big packets as there is less padding needed for 1492/1500 + a big overhead, compared to the amount for pppoa at 1500. Finding what your overhead is still be usefull if you want to run a smaller MTU. If you can get a cell count you should be able to find a data figure that changes from
2 to 3 cells and work it out from there. If you can't get a cell count then you may be able to notice a difference in IP throughput between MTU = X and MTU = X+1. 53 bytes per packet is noticable in pppstats.
If you can't see a difference then it maybe you are being throttled at IP level to below link speed already - I heard of some phone exchanges in the UK being like this.