[c-nsp] SLAs on a Gigabit Ethernet link

Adam Greene maillist at webjogger.net
Tue Sep 19 11:58:26 EDT 2006


Thanks to all those who replied, on or off list.

To summarize the feedback I received:
-    consider monitoring SLA thresholds utilizing dedicated equipment
hanging off 3560G's on each end of link
-    consider monitoring SLA thresholds on radios
-    be careful as radios may not (always) provide the expected 1Gbps
capacity
-    definitely make SLA compliance dependent on values measured by
service-provider infrastructure or service-provider monitoring devices (i.e.
do not depend on customer equipment to provide SLA compliance measurements)

I will reflect further on the feedback provided.

I am trying to decide at the moment between
1) simply telling the customer, "The SLA will be based on performance
measurements which I can provide (network availability, latency, and packet
loss). I can guarantee service uptime always, and I can guarantee latency
and packet loss at times of non-link saturation. But I can't provide you
with a guarantee that 1Gbps bandwidth (minus overhead and framing [x%]) will
always be available"
or
2) setting up dedicated *nix machines at either end which can be used to
push traffic through the link and show the customer, "look: the link can
pass 1Gbps (minus x%)" at any given moment (or periodically throughout the
day).

Just mulling things over ... perhaps I could also monitor performance
thresholds and stay below the terms of the SLA even during link congestion,
if I prioritize the monitoring traffic via some QoS mechanism (DSCP/CoS).

Finally, to answer a question which came up: this will be a dedicated point
to point link, utilized only by the customer.

Again, thanks.
Adam







----- Original Message ----- 
From: "Rolf Mendelsohn" <rolf-web at cyberops.biz>
To: <cisco-nsp at puck.nether.net>
Cc: "Adam Greene" <maillist at webjogger.net>
Sent: Tuesday, September 19, 2006 2:08 AM
Subject: Re: [c-nsp] SLAs on a Gigabit Ethernet link


> Hi Adam,
>
> I think your're looking / thinking about IP when you should focus on the
> radio's ;>).
>
> Firstly verify that they are full gigabit capacity (i.e. Gigabit ethernet
> capacity not Gigabit radio capacity).
>
> Secondly verify that the radios are actually full-duplex (most are not!).
>
> Lastly your assumtion that the link 'either passes traffic or doesn't' is
also
> incorrect, as you might experience interference (leading to packet loss),
or
> other related radio problems (perhaps changing the modulation from 256QAM
(in
> you case 1G) to 64QAM - perhaps 100M or 16QAM - 16M (just an example).
>
> So the best way of monitoring real performance and problems on the link
would
> be to poll the SNMP MIB's on you radio kit and not on your ciscos.
>
> HTH
>
> cheers
> /rolf
>
> On Monday 18 September 2006 22:23, Adam Greene wrote:
> > Hi,
> >
> > Perhaps this is asking a lot, but if anyone is able to point me in the
> > right direction or provide some good summary answers, I'd be grateful.
> >
> > I need to generate a Service Level Agreement for a customer that we're
> > going to be providing a point-to-point link for. On either end of the
link
> > will be a 3560G. The link will actually be wireless, via Gigabit radios,
> > but that's not so relevant. We'll be connecting two branches of a
Radiology
> > department. They plan to pass large MRI files between the two locations.
> >
> > First off, I am trying to figure out how to conceptualize the SLA.
Network
> > availability is a no-brainer (the link passes traffic or it doesn't).
> > However, the customer wants to be sure that the link is always
performing
> > well, and that the maximum bandwidth capacity of the link is always
> > available to him.
> >
> > I'm thinking of making use of Cisco's IP SLAs functionality to monitor
any
> > performance guarantees I may make to the customer.
> >
> > Some questions:
> > -    do the 3560G's even support IP SLAs? According to CCO, all IOS
> > supports it (except for some obscure exceptions), but I don't find any
> > reference to it in the 3560 configuration guide, and when I grope around
on
> > a 3750 in the lab, the switch seems not to support SLAs ("sh ip sla" is
an
> > unknown command; "ip sla ..." is an unknown configuration command)
> > -    it seems that the UDP jitter operation could be a good way to get
> > information about packet loss and latency on the link. However, I'm not
> > sure that obtaining these values will necessarily enable me to provide
the
> > customer with exactly what he needs. The customer needs to know that the
> > link supports up to 1Gbps of data traffic (minus any overhead) at all
> > times; if the customer himself is maxing out the link and I measure
latency
> > and packet loss, the results could appear poor. On the contrary, though,
> > the link would be doing exactly what it is supposed to be doing: passing
> > ~1Gbps traffic.
> >
> > Perhaps one way to approach monitoring of the SLA could be to
continuously
> > graph the utilization of the link and then associate the utilization
values
> > to the packet loss and latency at any given moment. But then it seems
that
> > I would need an algorithm of some kind to determine what acceptable
latency
> > and packet loss are on a 1Gbps link at varying traffic loads.
> >
> > Does anyone have any tips about how to approach these issues?
> >
> > Thanks,
> > Adam
>
>
>
>






More information about the cisco-nsp mailing list