[26012] in resnet
Re: Jumbo Frames
daemon@ATHENA.MIT.EDU (Buraglio, Nicholas D)
Tue Mar 1 13:49:53 2011
Content-Language: en-US
Content-Type: text/plain; charset="us-ascii"
Content-ID: <EB959BB7BD9CB54B99D953313A613F73@mx.uillinois.edu>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Message-ID: <D6BED195-FB1E-4F79-B538-C8036E220190@illinois.edu>
Date: Tue, 1 Mar 2011 18:47:57 +0000
Reply-To: Resnet Forum <RESNET-L@listserv.nd.edu>
From: "Buraglio, Nicholas D" <buraglio@illinois.edu>
To: RESNET-L@listserv.nd.edu
In-Reply-To: <AANLkTinnH6v-r14h-fuZzC0QeZrvvg3f9_16LeVSzSj+@mail.gmail.com>
I don't think of the jumbo support as adding speed, it's more like adding wider lanes to a highway rather than upping the speed limit. Larger trucks can bring more goods down the road.
I've found that it's useful for interconnect and for any network that may see large transfers of, for example, scientific data or research data sets. If your traffic patterns are mostly "many small packets" then it's not really a big win. I ran a broadband ISP for a while and never deployed jumbo frames there, because I was mostly dealing with what I called "consumer grade" traffic patterns.
Upon moving to the National Center for Supercomputing Applications in 2002, it was very apparent why large frames were used there. Large amounts of scientific data being moved around the LAN and across the WAN (WAN connectivity get a bit more hairy when you're dealing with long fat networks across large geographic regions). A properly tuned host with large frames enabled to another host with large frames enabled can hit the theoretical maximum transfer over a 10Gig network if all equipment has jumbo frames enabled and the hosts have 10G NICs. Thats probably outside of the scope of ResNet, but an interesting data point nonetheless.
On some smaller catalyst Cisco devices If GigE interfaces are configured to accept frames greater than the 10/100 interfaces, jumbo frames that ingress on a Gigabit Ethernet interface and egress on a 10/100 interface are dropped. I've never run into this issue personally.
nb
---
Nick Buraglio
Network Engineer
University of Illinois CITES / ICCN
GPG key 0x2E5B44F4
Phone: 217.689.4254
buraglio@illinois.edu
On Feb 24, 2011, at 2:13 PM, Dischiave, David wrote:
> We tested Jumbo's for backups on a per switch bases. It did not give us the additional speed we thought it would. If a Jumbo gets sent to a switch that does have have the higher MTU size set, it will drop the packet.
>
> On Wed, Feb 23, 2011 at 3:37 PM, Buraglio, Nicholas D <buraglio@illinois.edu> wrote:
> At UIUC we are all jumbo enabled within our backbone and throughout our RON. The only real issue I've seen in modern deployments is configuration related or user education issues, i.e. not having consistent jumbo paths or host tuning issues (for example, a 10G access port that is jumbo enabled). There are some older pieces of gear that don't do a full 9000 byte payload for their jumbo implementation as well, but I've not seen one in a long time.
>
> nb
>
>
> ---
> Nick Buraglio
> Network Engineer
> University of Illinois CITES / ICCN
> GPG key 0x2E5B44F4
> Phone: 217.689.4254
> buraglio@illinois.edu
>
> On Feb 23, 2011, at 11:27 AM, Randy Ethridge wrote:
>
> > I was curious as to how many people use or have configured support for jumbo frames on their network. Also did they run into any issues when they enabled jumbo frames?
> >
> >
> > Randy Ethridge
> > Network Engineer V
> > Information Services
> > Eastern Illinois University
> > rlethridge@eiu.edu
> >
> > Proud to say "I am EIU"
> >
> > EIU THINKS GREEN: Before printing this e-mail think if it is necessary
> >
> > ___________________________________________________ You are subscribed to the ResNet-L mailing list.
> > To subscribe, unsubscribe or search the archives, go to http://LISTSERV.ND.EDU/archives/resnet-l.html ___________________________________________________
>
> ___________________________________________________
> You are subscribed to the ResNet-L mailing list.
>
> To subscribe, unsubscribe or search the archives,
> go to http://LISTSERV.ND.EDU/archives/resnet-l.html
> ___________________________________________________
>
>
>
> --
> Thank you,
>
> Dave
>
___________________________________________________
You are subscribed to the ResNet-L mailing list.
To subscribe, unsubscribe or search the archives,
go to http://LISTSERV.ND.EDU/archives/resnet-l.html
___________________________________________________
