My Ongoing Cisco Rant

Okay, so if you didn't see my rant on VTP & GVRP, you may not understand my frustration. After spending a ton of money (at least it was compared to my budget) on a Cisco Catalyst 6509, I am starting to think I should have gone with another vendor. First it was GVRP. www.firewall.cx/ftopict-2693.html

%Begin Rant%
Then, tonight I decided to work on optimizing my gigabit traffic (aka trying to get even 500 mbps would be nice). I read and googled and read all about gigabit tweaks and recommendations. One of my first tests was setting up Jumbo Frames.

So, on two of my 5 year old servers, I set the MTU to 9000. Then, I log into my fancy 6509 to enable jumbo frames. We are using the WS-X6548-GE-TX switch module which is (should be obvious) a 48 port copper gigabit ethernet module. These things will run you about $10K a piece. I go:
[code:1]
lame6509(config-if)#mtu 9000
% Interface GigabitEthernet1/1 does not support user settable mtu.
lame6509(config-if)#ASLDJF;ALSJDFLAKJFDASL
[/code:1]
So that is my rant. I have to use the default mtu of 1500. Oh well, I guess that is life. I even called up cisco and they confirmed what I already discovered off google. Tough s**t!

I think I'll stop now. At least I wasn't the one who spec'd out this switch. If it ain't one thing, its another.
%End Rant%

Strange.. Setting a fixed mtu is a trivial feature of all NIC drivers in opensource OSs and I thought "jumbo frames" was a must to all gigabit products. Even most other Cisco series seem to support jumbo frames, however as they state clearly here WS-X6548-GE-TX does not. And using such a small segment size on gigabit links is a just a waste for nothing...
And.. 10.000$ you said? Oh dear :roll:

Strange.. Setting a fixed mtu is a trivial feature of all NIC drivers in opensource OSs and I thought "jumbo frames" was a must to all gigabit products. Even most other Cisco series seem to support jumbo frames, however as they state clearly here WS-X6548-GE-TX does not. And using such a small segment size on gigabit links is a just a waste for nothing...
And.. 10.000$ you said? Oh dear :roll:

My thoughts exactly. Some good news though. I used iperf (iperf homepage) to test my network performance and did get gigabit speeds using UDP packets. TCP speeds were <= 800mbps. It made me feel a little better but it still blows my mind that cisco doesn't support changing the MTU on these modules. It is actually one of the only devices in my data center that doesn't support MTU above 1500.

(the other device that I can not change the MTU size of... you guessed it... another cisco catalyst switch)

FYI, if you enable jumbo frames, every connected device needs to use jumbo frames. It seems like a good idea, but in practice it takes some planning in order to use it.

It can be enabled on either a global or per interface basis, of course as long as the platform supports it. I wonder if this is a hardware restriction or IOS related as far as lack of jumbo frames support? I was reading that some other 6500 modules limit the MTU size of jumbo frames to something like 8092 because of a limitation on it's ASICs.

My feeling is that it is because of a hardware limitation. I can set the MTU on VLAN's up to 18000+. I can set the MTU on my fiber ports to roughly 9000 and I can set the MTU globally to roughly 9000. However, my copper gigabit ports do not support user definable MTU sizes so I am stuck at 1500 for those ports.

Just because I was curious, I connected a couple of boxes on the fiber ports and set all the MTU's to 9000. I then used iperf to test throughput and found no noticable difference in network performance. I ran several tests with the MTU set to 1500 and 9000. My results were inconclusive because under different loads, there was a difference, but not one that stood out as the clear winner.

Personally, I feel there could be more testing because my testing could have been flawed. I did take great care in reproducing my results and ensuring I had controls in place.

Even if there is a performance gain, I don't feel it justifies the work involved in changing all the MTU's to a greater size. However, I speak as an admin for a corporate network of about 600 nodes. This may make sense for some University or Scientific labs that need all the bandwidth and availability possible...

Drew