[illumos-Developer] [PATCH] Jumboframes on intel 82573L

Garrett D'Amore garrett at nexenta.com
Thu Sep 9 09:55:39 PDT 2010


On Thu, 2010-09-09 at 18:46 +0200, Andreas Kohn wrote:
> On Thu, 2010-09-09 at 08:12 -0700, Garrett D'Amore wrote:
> > First, you need to file a bug in the bug database on this.
> 
> Ok, I created http://www.illumos.org/issues/178
> 
> > Second, I'm concerned about the choice for maximum MTU.  If the docs say
> > 9014 (which might be 9K plus an ethernet header), then asking for more
> > is clearly wrong.  Do the docs include the ethernet header itself as
> > part of the total count (14 bytes normally)?
> > 
> > 9014 as a choice is problematic for another reason: it means 9K plus an
> > ethernet header, without any room for VLAN tag. 9018 would be much
> > better.
> 
> Will verify this. Under the assumption that this indeed turns out to be
> a different number than MAXIMUM_MTU_9K, should I add another #define
> there?

Absolutely!

> 
> > So, before we accept this change, I'd like to see that you're able to
> > use all MTUs up to and beyond the size you've requested (9216), and that
> > you've verified this with testing.
> > 
> > That requires bumping the MTU up to the max value, and verifying that
> > e.g. ICMP ping of very large packets (say 32K or 64K) are passed
> > correctly and completely.  You need to change *both* the underlying
> > ethernet MTU *and* the IP MTU.
> 
> Thanks for the pointers how to test this properly, will verify.

Thanks.  If you have the ability, please also test VLANs with the large
frames.  You can do this by creating a back-to-back network.  That way
we can ensure that space is available for the 4 byte vlan tag as well.

	- Garrett
> 
> 
> As building the full sources is quite time consuming even with nightly
> -i, is there a quicker way to build and update just the driver for
> testing? 
> 
> 
> Regards,
> --
> Andreas
> 
> 
> 




More information about the Developer mailing list