9:05 p.m.
Hi all
How can I force the NIC to full duplex?
and change the mtu
where the file should I set it in boot time?
have you tried to force 10G full duplex?
Thank you
---------------------------------
All new Yahoo! Mail
---------------------------------
Get news delivered. Enjoy RSS feeds right on your Mail page.
9:18 p.m.
chloe K wrote:
Hi all
How can I force the NIC to full duplex?
and change the mtu
where the file should I set it in boot time?
have you tried to force 10G full duplex?
Thank you
You may want to check out ethtool as well as read
/usr/share/doc/initscripts-<version>/sysconfig.txt
Mikkel
--
Do not meddle in the affairs of dragons,
for thou art crunchy and taste good with Ketchup!
9:21 p.m.
On Fri, Jun 29, 2007 at 10:05:05PM -0400, chloe K wrote:
How can I force the NIC to full duplex?
ethtool -s eth0 speed 100 duplex full autoneg off
Note that the switch you're connected to will also have to be forced to full
duplex -- if one side is set to autonegotiate, the fallback is half duplex.
and change the mtu
ifconfig
where the file should I set it in boot time?
depends...
have you tried to force 10G full duplex?
Wait, hold on. What hardware do you have and what are you trying to do?
--
Matthew Miller mattdm(a)mattdm.org <http://mattdm.org/>
Boston University Linux ------> <http://linux.bu.edu/>
8:34 p.m.
Matthew Miller <mattdm(a)mattdm.org> wrote: On Fri, Jun 29, 2007 at 10:05:05PM
-0400, chloe K wrote:
How can I force the NIC to full duplex?
ethtool -s eth0 speed 100 duplex full autoneg off
Note that the switch you're connected to will also have to be forced to full
duplex -- if one side is set to autonegotiate, the fallback is half duplex.
Chloe] how can I put it in boot time? ls it in rc.local ?
and change the mtu
ifconfig
chloe] Do you know whether there is any network issue if we don't change MTU as same
as our provider? eg: our upstream provider router is using mtu 1600 but our end is 1500.
where the file should I set it in boot time?
depends...
chloe] can you give me example?
have you tried to force 10G full duplex?
Wait, hold on. What hardware do you have and what are you trying to do?
chloe] I will use the fedora as router. the NIC is "myrim 10G network card"
Thank you for your help
--
Matthew Miller mattdm(a)mattdm.org
Boston University Linux ------>
--
fedora-list mailing list
fedora-list(a)redhat.com
To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
---------------------------------
All new Yahoo! Mail -
---------------------------------
Get a sneak peak at messages with a handy reading pane.
10:54 a.m.
chloe] Do you know whether there is any network issue if we don't
change MTU
as same as our provider? eg: our upstream provider router is using mtu 1600
but our end is 1500.
If your MTU is lower than your upstream, you'll end up sending a few
more packets than you need to. (for 1500 vs 1600, about 1.066 times
to many)
If your MTU is higher than your upstream, then you'll fragment your
packets. When you try to send a single packet, it will get split into
two packets. (if you have 1700 vs 1600, the 1700 will be split into a
1600 and a 100 byte packet.) That doubles the amount of packet
overhead you have to transfer, and doubles the number of packets that
have to be received.
I'm not saying this authoritatively, but that's how I understand things to work.
11:09 a.m.
Michael H. Semcheski wrote:
> chloe] Do you know whether there is any network issue if we
don't
> change MTU
> as same as our provider? eg: our upstream provider router is using
> mtu 1600
> but our end is 1500.
If your MTU is lower than your upstream, you'll end up sending a few
more packets than you need to. (for 1500 vs 1600, about 1.066 times
to many)
If your MTU is higher than your upstream, then you'll fragment your
packets. When you try to send a single packet, it will get split into
two packets. (if you have 1700 vs 1600, the 1700 will be split into a
1600 and a 100 byte packet.) That doubles the amount of packet
overhead you have to transfer, and doubles the number of packets that
have to be received.
I'm not saying this authoritatively, but that's how I understand
things to work.
Just be careful that you have all the machines on the same network
(physical network/VLAN) all set to the same MTU otherwise bad things
happen. MTU sizes can change once you pass through a router. The
router will either fragment the packet providing the DF bit is not set
or send back an ICMP message stating that the MTU size has been exceeded
along with the correct/new MTU size to use.
Jeff
5:29 p.m.
Jeffrey Ross wrote:
Michael H. Semcheski wrote:
>> chloe] Do you know whether there is any network issue if we don't
>> change MTU
>> as same as our provider? eg: our upstream provider router is using
>> mtu 1600
>> but our end is 1500.
>
> If your MTU is lower than your upstream, you'll end up sending a few
> more packets than you need to. (for 1500 vs 1600, about 1.066 times
> to many)
>
> If your MTU is higher than your upstream, then you'll fragment your
> packets. When you try to send a single packet, it will get split into
> two packets. (if you have 1700 vs 1600, the 1700 will be split into a
> 1600 and a 100 byte packet.) That doubles the amount of packet
> overhead you have to transfer, and doubles the number of packets that
> have to be received.
>
> I'm not saying this authoritatively, but that's how I understand
> things to work.
>
Just be careful that you have all the machines on the same network
(physical network/VLAN) all set to the same MTU otherwise bad things
happen. MTU sizes can change once you pass through a router. The
router will either fragment the packet providing the DF bit is not set
or send back an ICMP message stating that the MTU size has been exceeded
along with the correct/new MTU size to use.
Jeff
Also note that some LAN's have a maximum MTU specified as part
of the physical design, and may not work (or work but with
poor performance) with a larger MTU. Traditional Ethernet
(the kind called CSMA/CD had a maximum MTU of 1500. Although
modern Ethernets have moved away from CSMA/CD, if that's an
Ethernet interface, I would advise leaving it at 1500.
5:51 p.m.
John DeDourek wrote:
>
> Just be careful that you have all the machines on the same network
> (physical network/VLAN) all set to the same MTU otherwise bad things
> happen. MTU sizes can change once you pass through a router. The
> router will either fragment the packet providing the DF bit is not
> set or send back an ICMP message stating that the MTU size has been
> exceeded along with the correct/new MTU size to use.
>
> Jeff
>
Also note that some LAN's have a maximum MTU specified as part
of the physical design, and may not work (or work but with
poor performance) with a larger MTU. Traditional Ethernet
(the kind called CSMA/CD had a maximum MTU of 1500. Although
modern Ethernets have moved away from CSMA/CD, if that's an
Ethernet interface, I would advise leaving it at 1500.
I don't fully agree (or disagree) however yes its true 10 and 100 Mb
ethernet use 1500 byte frames. However I disagree to a point that for
Gigabit ethernet you should also run an MTU 1500. Gigabit ethernet
allows for Jumbo frames or 9000 bytes (aka 9000 MTU) and there is
nothing wrong with using this MTU size. In fact you will see a
performance gain...at least on paper as your system may not actually be
able to take advantage of the lower overhead.
Also there are plenty of other (dying) network interfaces 16Meg"broken"
ring=8136, ATM=4470, FDDI=4470 that have MTU's larger than the standard
1500 any it may still be possible to take advantage of those links (on
internal networks)
There are advantages to moving away from the 1500 MTU to jumbo frames,
*HOWEVER* you must make certain that all the devices on that segment all
share a common MTU.
Jeff
6:10 p.m.
Jeffrey Ross wrote:
John DeDourek wrote:
>
>>
>> Just be careful that you have all the machines on the same network
>> (physical network/VLAN) all set to the same MTU otherwise bad things
>> happen. MTU sizes can change once you pass through a router. The
>> router will either fragment the packet providing the DF bit is not
>> set or send back an ICMP message stating that the MTU size has been
>> exceeded along with the correct/new MTU size to use.
>>
>> Jeff
>>
> Also note that some LAN's have a maximum MTU specified as part
> of the physical design, and may not work (or work but with
> poor performance) with a larger MTU. Traditional Ethernet
> (the kind called CSMA/CD had a maximum MTU of 1500. Although
> modern Ethernets have moved away from CSMA/CD, if that's an
> Ethernet interface, I would advise leaving it at 1500.
>
I don't fully agree (or disagree) however yes its true 10 and 100 Mb
ethernet use 1500 byte frames. However I disagree to a point that for
Gigabit ethernet you should also run an MTU 1500. Gigabit ethernet
allows for Jumbo frames or 9000 bytes (aka 9000 MTU) and there is
nothing wrong with using this MTU size. In fact you will see a
performance gain...at least on paper as your system may not actually be
able to take advantage of the lower overhead.
Also there are plenty of other (dying) network interfaces 16Meg"broken"
ring=8136, ATM=4470, FDDI=4470 that have MTU's larger than the standard
1500 any it may still be possible to take advantage of those links (on
internal networks)
There are advantages to moving away from the 1500 MTU to jumbo frames,
*HOWEVER* you must make certain that all the devices on that segment all
share a common MTU.
Jeff
Yes, I should have said 10/100 CSMA/CD Ethernets. Is GigE CSMA/CD?
I think it is switched. In any case, yes, my comment does not
apply to GigE.
6168
days inactive
6170
days old
8 comments
6 participants
participants (6)
-
chloe K -
Jeffrey Ross -
John DeDourek -
Matthew Miller -
Michael Semcheski -
Mikkel