MTU tuning?

Don Holmgren djholm@fndaub.fnal.gov
Fri Aug 14 12:44:17 1998


Thanks for the quick reply.  I tried the v1.02 version without
modification with no luck.  I also tried setting bit 5 in the
Cnfg register as shown in your patch, and still saw ttcp and
netperf hang.

I have no genuine need to boost the MTU, but was curious as to
any effects on performance.  I recently talked with a fellow from
KEK in Japan who'd used MTU=9000 on an Ultrasparc, running the
PacketEngines driver under Solaris.  He saw netperf numbers go
from about 180 Mbps with MTU of 1500 to about 280 Mbps with the
larger MTU.

Don Holmgren
Fermilab

> 
> On Thu, 13 Aug 1998, Don Holmgren wrote:
> 
> > Should I be able to successfully use MTU values other than
> > 1500?  Using the "mtu=" option when loading the yellowfin
> > driver works, as reported by 'ifconfig'.  However, subsequent
> > ttcp or netperf tests hang.  I'm using two GNIC's back-to-back,
> > and have tried MTU's of 2000 and 9000.
> 
> This is an untested "feature", and it only has a chance to work with a link
> between two machines, not through a repeater or switch.
> 
> It depends on the driver version.
> The earliest versions and most recent versions (e.g. v1.02) do work with a
> modified MTU.
> Somewhere in the middle I broke the receive buffer side -- it only allocated
> 1536 byte buffers.  The corrected versions use
>      yp->rx_buf_sz = (dev->mtu <= 1500 ? PKT_BUF_SZ : dev->mtu + 32);
>      dev_alloc_skb(yp->rx_buf_sz);
> rather than
> #define PKT_BUF_SZ		1536
>      dev_alloc_skb(PKT_BUF_SIZE);
>    
> Hmmm, looking at my notes, getting this work involved setting some other
> bit, which I now can't find in the documentation!  It may have been bit 5 in
> register 0xA0:
> 
> -	 	outw(0x01CF, ioaddr + Cnfg);
> +	 	outw(mtu > 1500 ? 0x01EF : 0x01CF, ioaddr + Cnfg);
> 
> Donald Becker					  becker@cesdis.gsfc.nasa.gov
> USRA-CESDIS, Center of Excellence in Space Data and Information Sciences.
> Code 930.5, Goddard Space Flight Center,  Greenbelt, MD.  20771
> 301-286-0882	     http://cesdis.gsfc.nasa.gov/people/becker/whoiam.html
> 

 | To unsubscribe, send mail to Majordomo@cesdis.gsfc.nasa.gov, and within the
 |  body of the mail, include only the text:
 |   unsubscribe this-list-name youraddress@wherever.org
 | You will be unsubscribed as speedily as possible.