Re: Estimated LTE Data Utilization in Failover Scenario

[Shawn Ritchie <me () shawnritchie com>]

> On Jul 31, 2019, at 11:03 AM, Blake Hudson <blake () ispn net> wrote:
>
> Matt Harris wrote on 7/31/2019 9:46 AM:
> > On Wed, Jul 31, 2019 at 9:21 AM Shaun Dombrosky <SDombrosky () blackfoot com <mailto:SDombrosky () blackfoot com>> 
> > wrote:
> > Good Morning,
> >
> > First time NANOG poster, apologies if I breach etiquette.
> >
> > Does anyone have any first-hand data on how much data a small-medium business (SMB) can expect to consume in a 
> > failover scenario over a 4G/LTE connection?  Retail, under 50 head count, using PoS, maybe cloud accounting 
> > software, general internet activity, 8 hour time period.  Wonder if anyone is using a Cradlepoint or SD-WAN solution 
> > that could pull a few quick numbers from a dashboard for me.  I haven’t had much luck in my searches.
> >
> > Appreciate any info anyone can provide.
> >
> > Thanks,
> >
> >
> > Hey Shaun,
> > I'd recommend pulling that data from the device normally facing their internet connection. Does it support netflow 
> > or even just basic snmp statistics that you could graph? Ostensibly the traffic level would be the same regardless 
> > of whether using an LTE backup connection or the primary internet connection unless you somehow prohibited certain 
> > traffic when on LTE. Ultimately though, your best bet is going to be to get real stats over the course of a couple 
> > of weeks and then you'll understand better the traffic patterns based on time of day, day of the week, etc, as well, 
> > as this is likely relevant. 
> >
> > Good luck! 
> 100% agree with Matt. Something also to keep in mind is the SMB's peak data rates. The primary (I assume ethernet) 
> uplink may have a sub 10ms latency and 100Mbps or greater data rate while the LTE connection is probably several 
> times slower in terms of bandwidth and latency. If designing a failover connection, customer expectations may need to 
> be managed: internet access may be up, but will be noticeably degraded when on LTE. A backup cable connection may be 
> better for VoIP or other latency/jitter sensitive applications and of course anything that relies on a static IP 
> (server, vpn, etc) will probably break if the primary connection is down. Would be a good idea to test the failover 
> connection during a few different time periods to gauge employee experience.
>
> —Blake

Yep. We sell solutions, both Cradlepoint and SD-WAN-based, and a big part of it is going over with the customer “you 
can’t just fail over all your regular traffic; pick biz-critical functions and deny everything else or you’re going to 
a) be very unhappy with speeds/performance and b) be EVEN MORE unhappy with the overage bill”. 

Get some data over a regular work week or so of their traffic, preferably with some flow data so you know what kinds of 
traffic/apps are consuming the bandwidth. Have the customer ID which of those flows would be critical if the primary 
connectivity died; size the cell plan appropriately or, if that can’t be done due to data caps, make sure needed 
tunnels for backoffice-type stuff will even work over your particular solution, etc... help them figure out what else 
can be dropped in an emergency. 

Other thing to consider is that almost all US cell plans have a pretty small data cap, even “unlimited”, and our 
testing shows that just backend Cradlepoint or SD-WAN chatter can add up to a GB or 2 a billing cycle; need to make 
sure your configs explicitly block any cellular usage unless the primary connection has gone completely down.

— 
Shawn