Hey Guys, A large set of my devices are Gigabit and Jumbo Frame capable. All of my IP phones and Printers are not however, they're stuck in the 100mbit age... Is it possible to set the MTU for a particular Port/VLAN? Maybe I'm not understanding everything... but I see the SG300 as a system with just a ton of NICs. Typically on a PC, you can set the MTU of the NIC to match the established L2 network MTU. An L3 device can then correct MTU mismatch by IP Fragmentation. Since my switches are in L3 mode, it seems like I should be able to set the MTU for a particular L2 VLAN. Thanks
There's no reason I can think of why setting jumbo frames across the board would not be OK even for those 100M ports. Enabling jumbo frames just permits those larger frames to be processed, so for hosts that don't send/receive jumbo frames there is no impact (and no benefit).
Remember L2 jumbo frames switching is not the same as L3 jumbo packet routing - and AFAICT the jumbo frame settings on these switches are only pertaining to layer 2 (frames) and not layer 3 (packets). There is no concept of fragmenting layer 2 frames, nor any concept around MTU - those are L3 functions.
So go ahead and enable it - I would be surprised if enabling jumbos had any negative impact on your phones.
Most if not all of Cisco's lower end products including the Catalyst 2000/3000 only allow per-switch level setting of jumbo frames, you need to look at the Cat 4000s and up to get the per-interface option.
First, thank you a ton for answering. There actually isn't a whole lot of info on this subject around, though it can make a huge difference in network performance. So I was under the understanding that all devices on your L2 network have to use the same MTU size.
Lets say I just have a flat L2 network with no internal L3 routing. Then, lets say my firewall emits a 9k ethernet frame with the destination being the MAC address of an AAstra phone. If the phone's MTU is set to 1500, how would it interpret the massive frame?
Going the other way, the phone would emit 1500 frames, but since that's less than the 9k limit on the firewall, it should work just fine.
In the case of your firewall sending a 9k sized layer 2 (as in, a 9k ethernet) frame through a switch which didn't support (or wasn't enabled) for jumbo frames, then the frame would hit the ethernet switch fabric and be dropped on the floor by the switch, if the frame exceeded the maximum MTU of that switch. That will create a real headache for you - layer 2 frame problems like that are a real pain in the #ss because you can for example ping across these switches, and browse some sites, but larger packets silently disappear into the ether, and a seemingly random selection of websites will become somewhat unbrowseable (or you may get the ads and the headline, but not the content for example).
Another good example of this is if you have Active Directory replication going on across a switch with a low MTU in the middle but end hosts using a high MTU, because you'll have these continual flapping of the replication processing, and timeouts logged but with no apparent loss in communications. Many people have lost lots of hair on problems like that
If it was a layer 3 MTU (as in, IP packet) mismatch, then the whole IP ICMP path discovery mechanism would kick in, and it's likely that communication would probably work, notwithstanding a firewall which could (but should not) be blocking this crucial ICMP traffic. You should try and avoid layer 3 mismatches as well though, because MTU that is smaller than the packets that need to traverse the link relies on IP ICMP unreachable packets signalling to the IP stacks on the endpoints to lower their MTU. Now in an ideal world this works, but there are still firewalls out there which block this stuff so it sometimes doesn't...
The good thing is that practically all switches by default have a layer 2 and 3 MTU of 1500, and this is fairly standard across hosts as well, so out of the box things just work. It's not ideal for a storage/replication environment though, where the higher MTU can give you higher throughput with less IP overhead.
So, the rules are:
- Higher layer 2 MTUs are better, there's nothing to lose by setting these high. It is a very good idea to consistently set this to the same value on all switches so that you don't have to keep track of what is set high and what isn't. Knowing you can do a 9000 byte ethernet frame across the board is good, even if you don't use it straight away.
- Higher layer 3 MTUs set on routers and hosts are OK but less well used, these only come into play when you are passing through a layer 3/subnet boundary, ie the IP packet is being routed.
- You ALWAYS need to have your layer 3 MTU equal to, or less than, your layer 2 MTU, otherwise you will end up in a world of pain.
- Normal networking only requires an MTU of 1500, but you will get better throughput out of storage and data replication environments if you can go higher between these hosts, on account of larger frames carrying more data per frame and thus fewer headers - and less work for the host to fragment the data into 1500 byte frames
- Usually it is a good idea to have all hosts on a VLAN using the same layer 3 MTU. It's not mandatory but it helps in terms of IP ICMP path discovery and is a good idea
So in summary - at layer 2, you need to get it right and make sure that your end-to-end path supports at /least/ the maximum MTU of your hosts, because there is no mechanism at layer 2 to deal with a mismatch. There's no real disadvantage to exceeding that minimum either so enabling jumbo frames is almost always OK.
At layer 3, things are a bit more flexible, and better at handling a mismatch so you can sometimes get away with more, but it's still not perfect. But overall, this is a slightly better situation to be in than the ethernet frames being dropped without trace :-)
Forgot L3 for a moment... Lets say all my L2 gear is Jumbo-frame enabled. If my phone is set to 1500 and my server is set to 9k... will they communicate bi-directionally without problems? Keep in mind, all of the intermediate L2 connects fully support Jumbo switching.
That should work OK (I have that same situation here). The two hosts will negotiate a mutually agreeable maximum "segment" size during TCP/IP negotiation which will work for both, and the IP packets will all be <=1500.
Sx550X, Sx350X, Sx250: PSE will Supply Power to Catalyst PSE Ports
May 31, 2016
June 5, 2017
Configure Remote Network Monitoring (RMON) Events Control Settings on a Switch through the Command Line Interface (CLI)
Remote Network Monitoring (RMON) was developed by the Internet Engineering Task Force (IETF) to support...