I've got a network with performance problems at GigE speed and would like to use jumbo packets. The problem is that there is some 100baseT gear on the network. What's the best way of using a Linux router to solve this given that I don't want to change IP addressing? I'm thinking of changing the addresses of printers (which I think includes all 100baseT gear) from 192.168.0.x to 192.168.1.x, adding all the old printer addresses to the GigE port on the Linux, and then using DNAT rules to forward the packets to the new printer addresses (which will be on a 100baseT switch on a different Ethernet port of the Linux box). That way I don't need to reconfigure everything that uses a printer. Is there a better way? -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
You definitely want to be using jumbo frames for gig ethernet; I advise using ethernet chipsets with hardware off-load as well, and ensure it's enabled at boot. I don't understand why you're worried about enabling the jumbo frames though. It doesn't break backwards compatibility. Your 100baseT stuff will continue to function fine. On 26 May 2014 17:38, Russell Coker <russell@coker.com.au> wrote:
I've got a network with performance problems at GigE speed and would like to use jumbo packets. The problem is that there is some 100baseT gear on the network.
What's the best way of using a Linux router to solve this given that I don't want to change IP addressing?
I'm thinking of changing the addresses of printers (which I think includes all 100baseT gear) from 192.168.0.x to 192.168.1.x, adding all the old printer addresses to the GigE port on the Linux, and then using DNAT rules to forward the packets to the new printer addresses (which will be on a 100baseT switch on a different Ethernet port of the Linux box). That way I don't need to reconfigure everything that uses a printer.
Is there a better way?
-- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
_______________________________________________ luv-main mailing list luv-main@luv.asn.au http://lists.luv.asn.au/listinfo/luv-main
-- Turning and turning in the widening gyre The falcon cannot hear the falconer Things fall apart; the center cannot hold Mere anarchy is loosed upon the world
On Mon, 26 May 2014 17:47:06 Toby Corkindale wrote:
You definitely want to be using jumbo frames for gig ethernet; I advise using ethernet chipsets with hardware off-load as well, and ensure it's enabled at boot.
Not always, it's a PITA to configure and most networks don't need it. My home network is running at 100baseT and I don't have problems with that speed.
I don't understand why you're worried about enabling the jumbo frames though. It doesn't break backwards compatibility. Your 100baseT stuff will continue to function fine.
Except that a switch will drop a jumbo packet destined for a non-jumbo port. So you can have situations where things work at low speed but break as soon as you send lots of data and get a larger TCP segment size. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
On 26 May 2014 18:07, Russell Coker <russell@coker.com.au> wrote:
On Mon, 26 May 2014 17:47:06 Toby Corkindale wrote:
I don't understand why you're worried about enabling the jumbo frames though. It doesn't break backwards compatibility. Your 100baseT stuff will continue to function fine.
Except that a switch will drop a jumbo packet destined for a non-jumbo port. So you can have situations where things work at low speed but break as soon as you send lots of data and get a larger TCP segment size.
I've never seen that in practice, and I've been running gigabit networks for a while. Rather than totally dropping the packet if the destination port doesn't support it, the switch should alert the sender that they must fragment their packets. Path MTU discovery. Although actually I think some (most?) switches instead just do the fragmentation themselves.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Mon, 26 May 2014 06:39:26 PM Toby Corkindale wrote:
Rather than totally dropping the packet if the destination port doesn't support it, the switch should alert the sender that they must fragment their packets. Path MTU discovery. Although actually I think some (most?) switches instead just do the fragmentation themselves.
That's routers, not switches. TCP will work on a subnet with hosts with different MTUs as you'll have the MSS taking care of that, but UDP won't know and so will do odd things. A long time ago, on a network far far away we put in (unmanaged) gigE switches for (then UDP) NFS traffic and chose switches that explicitly said they supported jumbo frames. Weird things started happening and we then realised that these switches were not doing jumbo frames as advertised. After a brief comic interlude with the vendor saying "turn on jumbo frames in the management interface" to which we replied "which part of 'unmanaged switch' did you not understand?" they ended up shipping out replacement ROMs for the switches and then it all worked. :-) All the best, Chris - -- Chris Samuel : http://www.csamuel.org/ : Melbourne, VIC -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.22 (GNU/Linux) iQEVAwUBU4SR0I1yjaOTJg85AQKT2Af/aZlBL/jUEhHc/2O+tQlFJaUkVqduEjDZ A1XFpOVkBf1pu9B/SGQN8/r4cGeBNOD5QWa8SvGL6lTq3Ho5f6ve2oMtaW1ijurN mzc8U4+yKktEy6alpuGN8tvMIpZxkfyLBODsAZLhdbJCHtk7OFYDUlczqUE2ho0U bciUpouCs1D3Fp/zBwUnGaIIu+yWfb98o+Aqco69wh83lHMxVkfBv9hATxwyZ5zp Q+Qom+hD9MIUdzRFLkLXjG3Lu9+fB6BWQLZt70FRLgTuuXSV9jyENeOgxQid4c+C 6T5iJNzWI4rgKXjsy4OlU/qEt6Z5uILbGOCWF3NLsPkB0DAHtvzLTg== =K0yB -----END PGP SIGNATURE-----
participants (3)
-
Chris Samuel -
Russell Coker -
Toby Corkindale