[c-nsp] Packet-level iSCSI debugging

[Mike Hale]

Nick:  We are not using Jumbo Frames or QoS yet, but we haven't seen
any indication of packet drops caused by saturation of the links.  The
hosts and storage are primarily plugged into the 2ks, and we are
seeing the issue across multiple ones.  It does span multiple LUNs,
and I believe they're spread among the two SPs.  The CPU of both SPs
is well below 50%.

Blake:  Do you know specifically which program?  We've got an OptiView
XG...thanks for reminding me.  I'll have to see if this sucker has the
tools we need.

Nick:  You're absolutely correct...thank you.  I'm not seeing any
drops on any interfaces, so if something's happening to those packets,
it's not being logged by the Nexus.

On Sun, Feb 2, 2014 at 9:03 AM, Nick Hilliard <nick at foobar.org> wrote:
> On 02/02/2014 01:41, Mike Hale wrote:
>> the utilization is well below 10gigs
>
> what you mean here is that "the utilization is well below 10gigs averaged
> over the sampling period".  Iscsi is sensitive to dropped packets, and it
> could be that you're dropping packets due to traffic bursts which are too
> short to see on your graph sampling period (300 seconds? most graphs use
> 300s by default).  Check out the dropped packet counts on all your iscsi
> ports and see what's happening there.  Even better, monitor the packet drop
> rate on your graphing system and build up a profile of what's happening.
>
> Nick
>



-- 
09 F9 11 02 9D 74 E3 5B D8 41 56 C5 63 56 88 C0