| [Other] MTU quandry! Greetings,
We have been clunking along with default settings for some time, but with the help of your web site and tools, I am trying to optimize our networks now. I have learned on this site that appropriate MTU settings are as follows:
Internet: 1500
Ethernet: 1492
VPN: 1400 (or 1300?)
I have also read that all PCs and routers should be set to the same MTU. Seems simple enough but I am administering a small office network (6 computers and 4 network printers) as well as a home network (2 computers and a network printer). Both networks have DSL connections and wireless routers. I use my laptop wirelessly on both networks, and just to keep things interesting, there is a VPN server at the office.
Our primary concern at the office is moving data around the LAN. At home it is the internet. VPN is less of an issue, and used primarily when I forget to bring files home or when traveling.
The office router was set to a MTU of 1492, the home router to 1500, and the PCs mostly at default values. I used the ping trial-and-error technique to establish an MTU at the office of 1492, but at home this technique yielded an MTU of 1006! (even with router set to 1500). Strangely, internet speed tests were faster at home with MTUs at 1500 than with MTUs at 1006.
My questions are as follows:
1) What MTUs would you recommend for my situation?
2) Why might packets fragment at 1006 at home?
3) If fragmentation occurs at 1006, why is performance better with a MTU of 1500 over the internet?
4) Can you really establish a different MTU for VPN transactions with a registry hack (I found a link to a Microsoft KB that implied this on this site.)
Thanks for your patience with my newbie questions!
Steve |
|
| MTU is generally set dynamically, however to optimise the network you can manually correct the issue...
Now Internet on DSL (PPPoE) is 1492MTU. Ethernet such as home LAN should be 1500MTU. VPN on the other hand needs to drop even more as the VPN overhead will take toll on the network and hence 1300 MTU...
Now, if your LAN is somehow using 1006MTU than there must be some cabling or other types of issue such as over subscription of the broadcast domain (i.e. your little hub/switch)
I hope this helps |
|
| Thanks ntWizard, that sort of helps, but I'm still not clear on what I should actually do. When you say that MTU is generally set dynamically, what do you mean exactly? Is there an "auto" setting or something similar? Both of our routers/wireless access points are Linksys and one has an "Auto" setting for MTU. Am I best off setting it to Auto and leaving all the PCs at their defaults? If they have been manually set using DrTCP or by editing the registry, what is the best way to allow them to be set dynamically?
If we manually set everything to 1492 is this likely to make things worse? Does any of this make any real difference or are we talking a percent of two of performance increase?
Thanks again, and sorry to respond with still more questions!
Steve |
|
| reply to stevemayman
When TCP/IP communicates with it needs to set certain things so that the traffic has no issues before it can proceed. example when you want to hit www.google.com your linksys tells your PC that 1492 is the right MTU for communicating with google. However, when the packet is moving to google that MTU may actually change by another device (such as a VPN) and drop the MTU to 1300. the user is not involved in this process and hence does not notice this issue and in reality this is really the case as every where you visit might have a different capability for example if you change your PCs MTU to 1492, than it might effect the local LAN speed. As an example Gigabit Ehernet can handle Jumbo Frames (MTU of 9216)..
Again, I would leave all the local LAN PCs alone as they communicate with each other... what I would do is to look at the cabling and make sure that there is no collision occuring as these collisions cause more problems than a miss guided MTU ...  |
|
| Thanks. That points me in the right direction. I guess your advice would be to set our routers to "Auto" and to leave our PCs at the default settings. For those that have already been manually set, will using DrTCP and deleting the MTU value essentially reset them to dynamically adjusting?
In your last paragraph you mentioned cabling issues and making sure there are no collisions. You don't mean the physical cables do you? I am guessing that this is beyond the scope of this forum, but is there a good place I could look to help me sort it out?
This may indeed be the problem because we have a Linksys WAP/4-port router at one end of the office connected to a Linksys 8-port switch at the other end of the office with a cat-5 cable. All the ports on each device are full. Things seem to work well most of the time, but the network freezes occasionally for 30 to 60 seconds. Sometimes we power cycle the WAP which gets things going again.
Any ideas?
Thanks again!
Steve |
|
| wow, great information... you see the 8 port switch should not see any collisions as they are full duplex(you have mentione that above) but the WAP uses a 4 port HUB... this is the reason why sometimes on a busy day broadcasts from multiple PCs can Freeze the network... to improve the situation is to add a switch and plug all the devices attached to the WAP hub on to the switch than by connecting the switch to the WAP will give you a smoother network reliability...
I hope that made sense |
|
| Yes, that made perfect sense. I would gladly buy a switch in hopes of fixing the problem. Unfortunately when I took another look at the router it turned that it is a switch after all. I guess there is no magic bullet for us, but thanks for your help!
Steve |
|
 cdruGo ColtsPremium,MVM
join:2003-05-14
Fort Wayne, IN
kudos:5
 ·Frontier FiOS
| reply to Da Geek Kid
said by Da Geek Kid:When TCP/IP communicates with it needs to set certain things so that the traffic has no issues before it can proceed. example when you want to hit www.google.com your linksys tells your PC that 1492 is the right MTU for communicating with google. However, when the packet is moving to google that MTU may actually change by another device (such as a VPN) and drop the MTU to 1300. the user is not involved in this process and hence does not notice this issue and in reality this is really the case as every where you visit might have a different capability for example if you change your PCs MTU to 1492, than it might effect the local LAN speed. As an example Gigabit Ehernet can handle Jumbo Frames (MTU of 9216).. The issue with MTU is that you want to get the maximum amount of data in each packet to minimize the effect of packet overhead. Each packet has a little bit of overhead. Straight TCP/IP over Ethernet has relatively little. PPPoE adds a little more (hence the 1492 instead of 1500). VPN has quite a bit more, etc. Normally the two machines will try to negotiate a happy number were the sending computer sends data that completely fills what the receiving computer (or some computer inbetween) will accept.
However in some cases there is a chance that a packet needs to get fragmented. For instance if the sending computer sends a 1500 byte packet but the path takes the packet across a PPPoE link in route to the receiving computer, it will get split up into a 1492 byte packet and a 8 byte packet. In the case of the 8 byte packet you are "paying" the overhead for another 1484 bytes of data, but you aren't actually sending it so that slice of bandwidth is gone.
Typically speaking you want to set your MTU as high as possible between you and your ISP and let the rest of everything handle itself automatically. If you are on PPPoE, set it at 1492. Most other DSL and cable connections are 1500. If you have something else, well it could be different.
--
Quis custodiet custodes ipsos? |
|
| reply to stevemayman
router does not change an existing MTU... The sender needs to resend the packet with the correct MTU size. If the packet is larger than the MTU. Router will trash the packet and send a notice to the sender to resend the packet with a MTU size that would allow the traffic to go thru |
|
| Thanks for the great explanation. So, in a perfect world, the router is set to auto (or 1500) so it does not become a bottleneck. Then the sending computer sends a 1500 byte packet and if it goes through, this continues maximizing your throughput.
If the packet has to travel along a path that has a lower MTU for one or more links...
1) The packets are split which doubles the transmission overhead. In the case above this overhead is just 8/1492 which doubles to 16/1492 so you would presumably only double the 0.5% overhead adding 0.5% to your max throughput speed. Not a big deal for a typical user like me.
or
2) The router asks the sending computer to resend the data with a smaller packet size that will fit through the pipe. The sending computer then splits the packet and you end up with the situation in #1 above, except that now each piece of data may have to be sent twice, which could presumably double the transmission time. A big deal!
It also sounds like the sending computer should "learn" that 1500 byte packets are too big and begin reducing the packet size until it no longer gets resend requests. In this case the situation is self-correcting and so there may be numerous re-transmissions at first, throughput is quickly maximized.
Soooooooo...
You need to set your MTU to a value below 1500 only if 1) You KNOW that there is a bottleneck in your transmission path (presumably because of your protocol or infrastructure) 2) either the auto-negotiation described above isn't working or you don't want the small amount of lost packets resulting from the auto-negotiation process.
For our 3 different requirements (Ethernet, TCP/IP, VPN) it seems like a no-brainer to let the transmissions auto-negotiate because we will use the highest throughput much of the time, and we don't want to limit that. It also sounds like the auto-negotiation has a small overhead (1% or so) which is well-worth the price of simplicity!
Does that sound like a relatively correct understanding?
Should I be worried about my 1006 byte limit at home? Is this a sign of a hardware problem? Based on my new understanding the worst-case is that OVERHEAD will increase by 33% due to the packet size but since overhead is small anyhow this shouldn't make a noticable difference anyhow.
Thanks again for all your help!
Steve |
|
|
|
 cacrollEventually, Prozac becomes normalPremium
join:2002-07-25
Martinez, CA | said by stevemayman:If the packet has to travel along a path that has a lower MTU for one or more links... 1) The packets are split which doubles the transmission overhead. In the case above this overhead is just 8/1492 which doubles to 16/1492 so you would presumably only double the 0.5% overhead adding 0.5% to your max throughput speed. Not a big deal for a typical user like me. or 2) The router asks the sending computer to resend the data with a smaller packet size that will fit through the pipe. The sending computer then splits the packet and you end up with the situation in #1 above, except that now each piece of data may have to be sent twice, which could presumably double the transmission time. A big deal!
Steve,
There is one additional possibility here.
3) Some servers absolutely can't handle fragmented transmissions (split packets). They drop fragmented packets, and you lose the connection. You see a white screen with no error, just no data.
»www.netheaven.com/pmtulist.html
--
Cheers,
Chuck
MS-MVP [Windows - Networking]
PChuck's Network |
|
| reply to stevemayman
number 2 is actually correct... the sending computer DOES learn about the accepted packet size for the specific destination. This in no means that the next packet headed to a different source would use the smaller MTU... All new sessions start with a default or a set features... |
|
| Great. I think I have finally got it. Thanks again to all!
Steve |
|
