Vista - MTU Size Query

Hi there,

I am trying to determine my optimum MTU Size on a PPPoA ADSL Connection - VC Based.

I have a Netgear DG834G v. 3 Router with latest UK V4.01.06 Firmware on Vista Ultimate x64 SP2 (SP2 stable as you like since March - btw)

When I Auto Detect Connection using Wizard in Router it always sets MTU in Router as 1458.

My ISP tell me to set MTU to 1432 for stability. As I type it's set to 1500 in both Router and Registry using a command I found somewhere:-

netsh interface ipv4 set subinterface "Local Area Connection" mtu=1500 store=persistent - then rebooting.

I must admit though it sometimes loses connection and when downloading test file from thinkbroadband :: Download Test Files the speed is up and down up and down. It's not much better on other MTUs.

Is the MTU Optimum size ISP, Router or OS specific? I understand the ping until fragmented test in Vista gives inaccurate results.

Ever since following the wiring diagrams at DSL ZoneUK - Improve your speed - Extension wiring and Disconnection help I've gone from 2 Meg to 8 Meg Broadband so thought may as well go the whole hog and sort this MTU thing out.

I must warn you that I've jigged around with various so called optimisers which may well have knocked some settings out although I have used

netsh winsock reset

and

netsh int ip reset

to try and even things up

help! lol

Short version first: leave it alone unless the ISP is making specific recommendations. Otherwise, messing with these settings is almost guaranteed to reduce internet speeds and increase the incidence of problems.

More detail: MTU is a function of a given network segment. For a standard ethernet LAN, an MTU of 1500 bytes means the overall packet size including all headers, payload, and trailers must be no more than 1500 bytes. The combination of the MAC, IP, and TCP headers generally takes up 40 of those bytes, depending on the TCP options in particular, so that the per-packet TCP payload size is reduced to 1460B - the "Maximum Segment Size" (MSS). If you artificially hard-wire the MTU to something very low, say 576B, approximately 3 times as many packets will be required for the same payload because the MSS is then only 536B. Extra overhead means less performance.

Path MTU (PMTU) is the least common denominator MTU value between two points A and B on a network. For example, if A is separated from B by 9 network segments connected by routers, each of those segments will have its own MTU and the PMTU will be the lowest number of them all. It would be very inefficient to continuously send 1500B packets across a 9-hop link, only to discover that the 7th segment has a lower MTU which necessitates packet fragmentation. That inefficiency is avoided by use of "PMTU discovery" mechanisms whose goal is to deduce the PMTU for a given path through the network.

When one or more additional protocols have to "ride" the same link, their headers and trailers (if any) all have to fit within the space previously used for payload. There is a slight decrease in efficiency - less payload per packet - but the gains stem from not having to continously keep (re)fragmenting all large packets in response to ICMP "fragment too big - chop it up" messages.

The network stack is self-tuning and administrative tweaks are only required in very specialised situations. If the ISP has issued instructions to modify the MTU, that's the right thing to do (they know their network). Otherwise, if you're doing this because some Joe on a web forum said it will speedz up ur internetz, consider the fact that they probably 14 and unqualified to be pitting their wits against a network stack


My advice would be to restore all network stack settings to their defaults first, to then make any specific changes provided in writing by the ISP, and finally to leave all of that stuff in peace - unless you've got specific goals or concerns.

Okely Dokely.

You obviously "know your onions" so I will do as you say.

Kind regards

Chris