[Atlas-anchors-pilot] HW specs: the choice of *two*

[Romeo Zwart]

On 12/10/12 03:11 , Aleksi Suhonen wrote:
> Hello,
> 
> On 10/05/2012 07:43 PM, Martin Stanislav wrote:
>> On Fri, Sep 28, 2012 at 02:13:05PM +0300, Aleksi Suhonen wrote:
>>> On 09/27/2012 09:49 PM, Romeo Zwart wrote:
>>>> The system proposed indeed includes an on-board Dual-port GE interface
>>>> (Broadcom).
> 
>>> OK, so the MTU issue with this hardware still persists? Broadcom drivers
>>> don't do jumbo frames reliably.
> 
>> Do the broadcom NIC drivers in general have problems delivering jumbo frames
>> or your experience is specific to some OS platform&  driver combination,
>> exceeeding particular frame size, any other feature in use (802.1Q tags)?
> 
> My experiences have been with Linux 2.6.3x and Linux 3.2.x with XEN and 
> 802.1Q tags. As long as the MTU settings are normal (untouched default 
> values, 1500+tags) everything works fine. But if I set MTU to 9000 the 
> driver will suddenly stop forwarding traffic after a random period of 
> time. It seems to work at first, but then fails when you're not looking.
> 
> The driver in question is bnx2 and the chip in question is:
> 
> 02:00.0 Ethernet controller: Broadcom Corporation NetXtreme II BCM5716 
> Gigabit Ethernet (rev 20)
> 
> This is a concern to me, because if we get to host an anchor and if 
> there will eventually be a K-root mirror on it, the K-root mirror will 
> need to connect to our peering VLAN which has MTU 9000.
> 
> So please test the proposed hardware platform with this in mind and let 
> me know if it's sturdy.

Thanks for this very useful information Alexi. We will jumbo frame
support in our testing.

Best regards,
Romeo