I have a customer that has an ESI phone system (proprietary VOIP).
The CPU is in Milwaukee (on Time Warner RR), they have four or five remote IP phones in Chicago (on Comcast, a connection sold as 12x2).
From time to time they complain of choppy/stuttering voices. Happens in the afternoon a lot.
Digging into the issue, it seems Comcast has this “Network Management” tool where they will throttle users if they consume “too much” bandwidth.
Now, this customer doesn’t use much bandwidth. Maybe 1.5 to 2GB down and a small fraction of that up, per day.
But some Comcast customers have posted that Comcast gets congested and they throttle back EVERYONE.
At first I thought the 12x2 connection for $60/month was a bargain. But it looks like Comcast way oversells their bandwidth, and anyone doing VOIP pays the piper in reduced call quality.
I’d even be fine with this if Comcast would set some minimum standard, like “that 12x2 will not be reduced below 8x1.” At least we could look at the menu of services, buy the package that meets our minimum requirement, and configure things appropriately.
Does anyone have any Comcast experience? Is there any way to gauge what they do to connections? Any sort of “rules of thumb” would be appreciated.
Comcast is not in our market so I have not run into them.
We are a Time Warner Business partner I can sell a BCE, a point to point over DOCSIS and deliver full QoS. We can do the same with Fiber. Each customer we add TWCBC just creates another VLAN on our Gig uplink.
We also have a Time Warner BGP peer, if you are on TWC Northeast we can deliver SIP/IAX trunks without leaving the TWC Network.
They are compelling products and a great business partner.
The HSI offering on Comcast has 2 DOCSIS class mappings: Priority Best Effort (PBE) and Best Effort (BE). The higher DOCSIS class mappings are reserved for traffic like their Comcast Digital Voice product.
By default, every cable modem starts out in the PBE class. This remains true until the cable head-end detects a near congestion state on a segment of the cable network. The thresholds are available in the docs linked from the site above, but I believe the threshold is 80% utilization. Once that trigger occurs, the system identifies connections that have been using more than some percentage of their subscribed capacity (ie > 50% or 60%) for the last 15 minutes, and those connections have their traffic re-marked into the BE class.
It’s not really a throttle, because even though the traffic is re-marked into a lower class, if the channel never becomes congested then no traffic gets dropped. What this system does is that if the shared link reaches 100% utilization, the heaviest users get their traffic dropped first. The reality of the situation is that when a link is running at 100% utilization some traffic is getting dropped no matter what, this system just attempts to make which traffic gets dropped a little more “fair.”
Business class on Comcast has its own support tier, and same-day service call escalation. While there is no SLA, I’m not sure how much I’d worry about that. SLAs aren’t a guarantee that something is going to stay up, it’s a mechanism for financial compensation should services be unavailable. Paying more money for an SLA-based service just so you can get monetary compensation for downtime seems self-defeating.
Your post almost had me convinced Comcast’s “Network Management” was the greatest thing since sliced bread.
Just thinking aloud here…
If I buy a 50/10 connection and use it at 50/10 for more than 15 minutes when a bunch of other people in my “neighborhood” are using their connections intensively, I might find my consistency of service goes to heck.
If I’m doing VOIP and my CoS falls apart, the connection may as well go down.
So if I need to maintain consistency of service, I need to pay for 50/10, but only use 25/5.
And that would even be sort of okay.
But from my experience in a crowded market like Chicago, even that isn’t a guarantee (as my client never uses more than one-tenth of their bandwidth and they still find consistency of service falls apart most afternoons).
It seems to me that ideas like “Network Management” are designed to punish customers for using what you’ve sold them. The idea is to promise a lot, deliver a little, and avoid costly upgrades to infrastructure.
I suppose this will be the model for Comcast and many others going forward.
Question (for anyone): How many 50/10 connections (running full-tilt) does it take to saturate a cable segment?
Have you or your client opened a ticket with Comcast?
I get that it’s vogue now to rip on ISPs as being deplorable entities that exist solely to screw customers. While a lot of that ire is deserved, the numbers would tend to suggest a few ISPs are a little better at delivering a reasonable approximation of what their marketing departments have promised.
I bring these up only because they suggest that maybe Comcast doesn’t have systemic measures to screw their customers, ergo it might be worthwhile to report a performance problem because it would indeed be an anomaly.
As to how many 50/10 customer can saturate a cable segment, it depends on how many bonded channels are built out on that particular segment. Higher bitrates are achieved by bonding together multiple 38mbps channels, which can vary from 2 up to 8 bonded channels in current deployments.
First, many types of coaxial cable, it all depends on the frequency you are going to push through it. For the sake of this discussion let’s limit ourselves to the typical RG-6 used for wiring cable TV drops.
Not as easy an answer as you may think. Technically allocated “jumboband” frequencies end at channel 158, or a hair over 1Ghz. Uncoded data bandwidth would be around 1Gb/sec. However most North American providers are now modulating beyond 2Ghz. Remember this bandwidth is shared with TV, phone, management and other traffic.
DOCSIS 3 Max’s out around between 30-50Mb per channel. Consumer stuff is all single channel so that is why the DOCSIS offerings top out at 50M using 64-QAM or S-CDMA coding.
If you want a deeper of information coding understanding the wiki page on Shannon’s law is outstanding:
Technically, the DOCSIS downstream channels are 38mbps (effective throughput) per 6MHz of frequency space. DOCSIS 3.0 achieves higher data rates by bonding multiple channels together. Most systems are doing at least 4 channel bonding, with some markets now seeing 8 channel deployments on various providers.