Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[boost]

Boost,
That post should be a 'Sticky' quick OSI 7 layer to mtu explanation.
Nicely done.

Question:
What is the real impact of reducing your mtu ?

Also many people use Jumbo frames, where does this fit in ?
(Add this to the post perhaps?)

TIA

Send from LG G3 via Tapatalk (Typos & bad formatting are free)

Thanks! I think we all need to be on the same page if nothing else. I've no idea idea what the impact is but considering you can hit gigabit speeds with a 1500 byte MTU, I think shaving a few bytes off it should be OK for 80Mb

[boost]

ejs, some good info there, thanks. I'll have a proper squint when I'm home

You may know better than me but I've only seen MSS rewritten for tunnel endpoints as a last resort. Having to inspect all new TCP connections is not without it's own resource penalty. Also, what about UDP? Hello blackhole?

I suppose all we can do is experiment and see what works best for us!  

[Alex Atkin UK]

Like I said, I have my whole LAN set to 1480 MTU and it doesn't appear to cause any performance issues at Gigabit speeds.  Sure it will use more CPU power, but with modern hardware that is pretty negligible.

I believe it can be useful for gaming, as the more data you can send in one packet, the less latency you have.

Obviously sending the same amount of data in 1500 chunks uses less router resources than with a smaller MTU.  But if your router is running THAT close to capacity you should be upgrading it anyway as its likely your NAT performance is being impacted.

I have been using an Atom as a router for a while so that I wouldn't have to worry about anything being a resource hog.  I was sick of MIPS routers being underspecified for QoS and that if you want to look at the connection table it HAMMERS the CPU, less of the problem when you have dual-cores.

[ejs]

Since the firewall (and NAT) in a router will be keeping track of the state of all connections, so that it knows what incoming packets to accept based on if they are part of or related to an existing connection, I suppose it's not too much extra work to modify a relatively small number of packets.

Yes, the MSS altering is only applicable to TCP, it can't help with UDP. But then I'm not sure setting yourself a lower MTU is going to help if a remote server sends you a 1500 byte UDP packet.

A smaller packet size will be very slightly less efficient, since a slightly greater proportion of your bandwidth will be carrying TCP/IP headers, but the difference is so small it may be too small to reliably measure any download speed difference let alone notice it.

[Chrysalis]

its not down to the smaller packet as such but rather a non standard MTU has to be negotiated on every connection.

[ejs]

20 byte IP header and 20 byte TCP header, overheads:

40/1500 = 2.67%
40/1480 = 2.70%

The remote server may use a lower packet size anyway, Google servers often do.

[Alex Atkin UK]

Well as a large percentage of users WILL be DSL customers, if that "negotiation" overhead was actually a problem it would make sense to choose something lower as standard.

[iMx]

Not true generally, only true for PPPoE.  PPPoA does not have the same 'problem' - this allows the full 1500.  The standard comes from ethernet, not DSL/broadband.

[Alex Atkin UK]

PPPoA is only really common over here and thanks to VDSL is gradually being phased out.  PPPoE is far more common worldwide.