Submit Hint Search The Forums LinksStatsPollsHeadlinesRSS
14,000 hints and counting!

Click here to return to the 'Optimising MTU setting in PPPoE ADSL Connections' hint
The following comments are owned by whoever posted them. This site is not responsible for what they say.
Optimising MTU setting in PPPoE ADSL Connections
Authored by: cahaddras on Aug 28, '05 07:42:39AM

With ADSL it is not always a matter of finding the highest MTU that works. All packets are transferred as multiples of 48 bytes over the link to the exchange, padding if necessary. The padding wastes throughput, so the trick is to set the MTU at the highest value that results in an exact 48 byte multiple, allowing for the protocol overhead. With a 34 byte PPPoE protocol overhead, 1454 MTU gives a 1488 byte packet which is exactly 31 * 48. 1492 MTU gives a 1526 byte packet, which is 32*48 including 10 bytes padding. If you do the math on an 8Mb connection, that gives a real data throughput of 856kByte/sec for 1492 MTU, and 875kBytes/sec for 1454 MTU.

Note that this only applies if the ADSL link speed itself is limiting. This is always the case for 8Mb, but for anything less than 8Mb some ISPs aim to set a higher link speed than needed and then cap the data throughput. In this situation the wasted pad bytes are just soaked up by the link speed headroom and make no difference on the data throughput. On the other hand if your distance from the exchange is forcing a lower link speed then tuning the MTU in this way will help get the best out of the reduced throughput.

[ Reply to This | # ]
Optimising MTU setting in PPPoE ADSL Connections
Authored by: Gort on Sep 10, '05 01:31:32PM

I'm not sure I follow the math here. Each 1492 packet will be 32 48-byte chunks, for a total 1536 bytes, of which 34 are overhead, 10 padding. Each 1454 packet will be 31 chunks, for a total of 1488 bytes, of which 34 are overhead. So every 47616 bytes sent will be 31 of the 1492 packets or 32 of the 1454 packets, and

31 * 1492 = 46252
32 * 1454 = 46528

which is a smaller difference than the 875:856 suggested in the article.

(I'm not qualified to comment on the substance; but when I tried to work through the arithmetic, I ended up off your numbers).

[ Reply to This | # ]