I got in this deal: http://www.hardforum.com/showthread.php?t=950009&highlight=switch and picked up 5 switches. I'm currently using the switches throughout my house, but I have no gigabit NICs. I will be buying the gigabit NICs this week, although I have some equipment that I can't upgrade to gigabit (mainly my router, some satelite equipment and a small server I have). Will there be any problems communicating with the 10/100 stuff if I enable all the gigabit nics to Jumbo frames? From what I understand, the gigabit switches are backwards compatible and will do the necessary changes to send the jumbo frame packets to the 10/100 equipment or is there more to it than just plugging in my old equipment to the switch?
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Upgrading to Gigabit, anything I should watch out for?
- Thread starter Time2Kill
- Start date
That's a good question. For the switch to reassemble or expand packets that the wrong size, it would have to know if the device attached to a given port supports jumbo or not. If this switch isn't managed, than I don't know how it would decide if a given port should be getting jumbo frames or not.
For a managed switch, you just set the max framesize for each port. The switch does the rest.
For a managed switch, you just set the max framesize for each port. The switch does the rest.
Malk-a-mite
[H]ard|Gawd
- Joined
- Feb 16, 2002
- Messages
- 2,023
Cranky fast posting Malk-a-mite continues after the quoted part below.
There are some switches that handle layer 2/3/4 stuff but at that point you are looking at things like the Cisco Catalyst 4000 series. mikeblas - if you have a look at the switch the OP is talking about it's a 50 dollar piece of plastic. So, yes mikeblas is right in that some switches can handle higher layer protocols and can help decide packet sizing... this just isn't one of those switches.
Orinthical's post in the Hotdeals forum is still the best advice you have for the network you are putting together. The whitepaper I link to below covers similar problems of putting gigE systems on the line with systems expecting 10/100.
Original ranting post continues from here
---------------------------------
Hold up.............
For a switch to do what?
Switches don't do anything other than pass traffic, they don't reassemable, they don't break down, they just pass it. It is up to the Data Link layer on the sending and recievnig ends of the connections to deal with packet and queue sizes, not the switch.
For a good explanation of the whole jumbo frame concept have a quick look at this:
http://searchnetworking.techtarget.com/searchNetworking/downloads/whitepaper_jumbomtu_sep2003.pdf
Edit:
Time2Kill - listen to Orinthical, he knows what he's talking about in this post:
http://www.hardforum.com/showpost.php?p=1028632151&postcount=74
There are some switches that handle layer 2/3/4 stuff but at that point you are looking at things like the Cisco Catalyst 4000 series. mikeblas - if you have a look at the switch the OP is talking about it's a 50 dollar piece of plastic. So, yes mikeblas is right in that some switches can handle higher layer protocols and can help decide packet sizing... this just isn't one of those switches.
Orinthical's post in the Hotdeals forum is still the best advice you have for the network you are putting together. The whitepaper I link to below covers similar problems of putting gigE systems on the line with systems expecting 10/100.
Original ranting post continues from here
---------------------------------
mikeblas said:
Hold up.............
For a switch to do what?
Switches don't do anything other than pass traffic, they don't reassemable, they don't break down, they just pass it. It is up to the Data Link layer on the sending and recievnig ends of the connections to deal with packet and queue sizes, not the switch.
For a good explanation of the whole jumbo frame concept have a quick look at this:
http://searchnetworking.techtarget.com/searchNetworking/downloads/whitepaper_jumbomtu_sep2003.pdf
Edit:
Time2Kill - listen to Orinthical, he knows what he's talking about in this post:
http://www.hardforum.com/showpost.php?p=1028632151&postcount=74
The switch that brought you this post is set to give a 9600-byte packets on the ports where I've allowed it to, and 1518-byte packets elsewhere. I can verify that it's accepting larger packets (on the ports so configured) and sending smaller packets (on ports so configured) using a network sniffer.
mikeblas said:
So you're saying that a home appliance switch is fragmenting the packets? lol, that's a layer 3 function dude, you'd need a router on a stick configuration with seperate vlans for that to work, I believe. Unless you had a layer 3 switch, fragmentation is a portion of the IP protocol, layer 3, layer 2 is packet switching based on MAC, it doesn't fragment or have the capability to fragment. Plus, that switch you have says it uses store and forward, so I don't know why they'd use that instead of cut-through, seems like a waste to be looking for giants witha 9600 byte frame...
Okay, let's say you win.
But even with that stipulation, how do you explain what I'm observing, then? I can't understand why the frames on one side of the swtich would be larger, while on the other side the same data would appear in a smaller frame.
Have you used this "appliance"? What did you observe it doing in your setup?
But even with that stipulation, how do you explain what I'm observing, then? I can't understand why the frames on one side of the swtich would be larger, while on the other side the same data would appear in a smaller frame.
Have you used this "appliance"? What did you observe it doing in your setup?
Rabidfox said:got me tho
Indeed, I do have you.
What I'm telling you is that I observe large packets being torn into smaller packets by the switch. I've used both Ethereal and Network Monitor to observe this, and that's exactly what is happening.
I never see the switch assembling smaller packets into larger ones, but it certainly does retransmit larger packets in groups of smaller ones.
How does it work, otherwise? Is there a negotiation through the switch that establishes the MDU? Was I just unlucky in not witnessing it?
MTU is determined by the IP protocol, so I don't see how a switch is fragmenting the packet into smaller frames, it would have to replicate the IP header and trailer every time it broke downt he packet. Try setting up ethereal on both ends, filtering the output with grep, and transmitting a single large packet (ping comes to mind) and see if it's it framgmented on the other side. Do this from the gige to 100e, then gige to gige. I'd be interested to see whats happening. There isn't a negotiation with the switch that determines MTU, for ethernet it defaults at 1500+overhead(1518). You have to manually set it for larger MTU's. The MTU is set by the host machine, regardless.
You're right, it is simpler, but it's still a layer 3 job, not a data link layer. That switch that you linked me was just a data link layer device, unless you're using a router on a stick with vlans or they're on 2 subnets be routed through a router, I don't think it can fragment packets. Were the devices (gige and 100e) on the same subnet?
UnrealRage
Limp Gawd
- Joined
- Sep 26, 2004
- Messages
- 155
Could be possible that the switch isn't fragmenting anything. When a computer that can use jumbo frames attempts to communicate with a computer of lower MTU the Receiver may be alerting the sender that it cannot accept that large of a MTU. Maybe then they agree upon a identical MTU that they both support prior to transmitting. At that point the sender may be backing off and agreeing to send a lower MTU to the receiver.
Setting the MTU on switch interfaces is definitely important for determining the lowest path MTU. Jumbo frame enabled hosts should have an identical MTU value configured on the switch interface. If the switch isn't configured correctly, you may then see MTU conflicts because the two computers agree upon 9000 bytes but the frames are dropped when it tries to exit/enter an interface only configured for default MTU.
Just a thought...
Setting the MTU on switch interfaces is definitely important for determining the lowest path MTU. Jumbo frame enabled hosts should have an identical MTU value configured on the switch interface. If the switch isn't configured correctly, you may then see MTU conflicts because the two computers agree upon 9000 bytes but the frames are dropped when it tries to exit/enter an interface only configured for default MTU.
Just a thought...
Malk-a-mite said:
What I'm planning on is taking a WinXP Pro box, sticking a 10/100 nic and a gigabit nic in it. Bridge the connection, set the packet size on the 10/100 nic to 1500 and enable Jumbo frames on the gigabit nic. Should there be any problems with that.
Orinthical mentioned in the other post that a setup like that would reduce my throughput if I use WinXP, but I only need a steady rate of about 500KB/s as I'm only supporting a T1 line through and some other small stuff on the 10/100 side. I currently see a pretty steady 9,000 to 10,000KB/s when testing over my 10/100 network right now.
I plan on eventually making the switch to Linux, but right now I just need to get everything running
Malk-a-mite
[H]ard|Gawd
- Joined
- Feb 16, 2002
- Messages
- 2,023
Time2Kill said:
If time is a problem - run with the WinXP bridged connection. When you have the time look to migrating to a more dedicated *nix box that is designed for the task.
Yep.Rabidfox said:
Let me make it clear that I'm not saying you're wrong -- I'm just saying it doesn't match what I'm observing. If there's some alternate explanation and I end up with a better understanding, that's awesome.
mikeblas said:
Hrmm, I spoke with a co worker who is a CCNP and has finished the written portion of the CCIE, he said that alot of cisco/managed equipment can do it, but he wouldn't know about appliance switches. It might be built into the firmware so people couldn't screw it up, thats a nice feature