[ale] network question

[James P. Kinney III]

That's what I'm seeing on further digging.

I'll have to modify my design to isolate my server to storage network
from the server to client network.

On Sun, 2007-08-26 at 23:54 -0400, James Taylor wrote:
> I've been working with optimization for iSCSI lately, and jumbo frames are big factor for it.
> Everything that I've seen or heard indicates that if any device on switch doesn't support jumbo frames, they get disabled.
> 
> Some of the switches support setting up jumbo frames in isolated vlans or even ports on the same switch, but any device that doesn't support jumbo frames will disable it for the rest of the devices on the same network.
> 
> So far the only devices I've worked with have been current generation Extreme switches
> When we enabled jumbo frames for the linux boxes, the storage throughput increased about 30%.
> 
> -jt
> 
> 
> James Taylor
> The East Cobb Group, Inc.
> 678-697-9420
> james.taylor at eastcobbgroup.com
> http://www.eastcobbgroup.com
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> >>> "James P. Kinney III" <jkinney at localnetsolutions.com> 08/26/07 10:43 PM >>> 
> If I have a server with a gig NIC connected to a switch with a gig NIC I
> can turn on jumbo frames on both and have each tcp transaction be more
> payload and less overhead.
> 
> OK
> 
> So if there is another switch one more hop away from my server and it
> also has jumbo frames on the traffic continues is a more efficient more
> from the server down to the second switch.
> 
> But jumbo frames are not supported by 100Mb NICs ( at least none I've
> seen) so jumbo frames won't work if the final data recipient doesn't
> support them. I'm guessing that the result of this will be an extra
> burden on each transmission setup to establish what the little one can
> handle. Not a good thing as I'm trying to squeeze the maximum bandwidth
> and minimum latency out of my network (tubes for the unknowing :).
> 
> So here's the final scenario: Same server and switches as before but add
> the following twist - the server has 4 gig NICs bonded to do 802.3ad
> (link aggregation) so they can do basically ethernet striping across 4
> ports so maximized bandwidth between AoE hard drive stack (w/ a 10G
> connection). But those same 4 wire bonds also talk down the switch
> connections to the downstream little devices. 
> 
> So the Linux question of the day is: is it possible to have a variable
> frame size based on destination or is the downsizing process of the
> frame rate so insignificant it won't matter? Or will the last switch in
> the chain handle the frame size conversion (SMC SMC8748L2
> 10/100/1000Mbps Managed Layer 2 Switch - still digging through the 480
> page users manual).
> 
> -- 
> James P. Kinney III          
> CEO & Director of Engineering 
> Local Net Solutions,LLC        
> 770-493-8244                    
> http://www.localnetsolutions.com
> 
> GPG ID: 829C6CA7 James P. Kinney III (M.S. Physics)
> <jkinney at localnetsolutions.com>
> Fingerprint = 3C9E 6366 54FC A3FE BA4D 0659 6190 ADC3 829C 6CA7
> 
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://www.ale.org/mailman/listinfo/ale
> 
-- 
James P. Kinney III          
CEO & Director of Engineering 
Local Net Solutions,LLC        
770-493-8244                    
http://www.localnetsolutions.com

GPG ID: 829C6CA7 James P. Kinney III (M.S. Physics)
<jkinney at localnetsolutions.com>
Fingerprint = 3C9E 6366 54FC A3FE BA4D 0659 6190 ADC3 829C 6CA7
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part