Web Hosting Talk : Web Hosting Main Forums : Colocation and Data Centers : Worth paying extra for no overselling?

[USSHC_Colo]

It's simply common knowledge that most providers at every level oversell on bandwidth. I've always avoided that but while looking at our new fiber routes and additional providers it keeps being brought up that we should simply oversell. I'm interested in thoughts on this one...

Say, would a $700 100 mb port (totally unrestricted and not oversold) be more valuable to you as a client than a $500 100 mb port that is likely oversold?

Curious to see what the market is really after...

Thoughts?

[InnovusHost]

Is this a 100mbit commit on a 100mbit port?

[USSHC_Colo]

Yes, let's say it's an unrestricted 100 mbit port - you could ramp it up as far as you want or use it as little as you want with no penalty.

[InnovusHost]

I think that would justify the premium. $7/mbit isn't a bad price to begin with. Would you consider selling that commit on a gigabit link? If I were looking to commit to 100mbit I'd like to see a large port to ensure that I am able to hit 100mbit without issues.

[Spudstr]

Everyone oversells and if you don't, your lying. _or_ you don't have the customers to fill your minimum commit.

[tulix]

I think from hoster point of view there is at least 2 levels of overselling:
1. when you know that you oversell and
2. customers are not using all their commitments and if suddenly all of them will start - then it will become oversell.

I think when prices for bandwidth are too low to be true it is or low quality bandwidth or number 1 above.

Some companies with large number of (usually) dedicated servers based on number 2 item pushing marketing campaign of selling bandwidth for a very lo price. They are counting based on their previous experience that nobody will use their commits, and, even if they are loosing money on some that do use their commits, - it is just a marketing cost for them.

Not sure if I've answered on your question, but at least I am sharing with my experience from hoster side point of view.

[RyanD]

I think the only people concerned with 'overselling' would be those in the general WHT marketplace. WHT is not the end-all-be-all of the hosting industry and these themes of rock-bottom pricing, overselling, low margins, etc generally do not apply to the rest of the hosting services industry as a whole.

If you are able to leverage the large size of your network with existing bandwidth usage and peering arrangements it is easy to sell a non oversold 100mbps solution at $5/mbps with a top notch network. It is pretty common these days to get your Tier-1 carriers at prices significantly below $5 (no I am not counting Cogent, they are not quality and their silly $1/meg pricing does not influence this)

I would agree that the overall 'oversold' trend in this industry is damaging and to the WHT market you may be able to upsell a non-oversold product but keep in mind the driving trend behind WHT buyers is rock-bottom pricing anyway and they often do not care about oversold or not.

I'll go on the record stating we DO NOT oversell and are not over-committed hoping customer's do not use their bandwidth. I would love to see other providers state that they run their network in the same way.

[USSHC_Colo]

We don't oversell, and I'm very happy about that. That's how I'd like to continue in the future. We have excess capacity on the circuits we pay for and it works out just fine. As we put in additional fiber runs I'd like to keep it that way. I don't ever want to be in a position where we have more commit than capacity - It would only be a headache for everyone involved.

It's hard to gauge the market. WHT does have that feeling of looking for the lowest price, but there are definitely people with a lot of experience that know you frequently get what you pay for.

[tulix]

I know what Ryan is talking about - WHT is a specific market, so why low end vendors, like Mzima, HE, Cogent are so popular in this market place. Top notch network is different for different markets - again, in our case we can't come even close to $5 per Mb/s, but we are talking about just different quality/markets.

Among popular vendors at WHT is Cogent that we are using in our network - interesting that we've got that vendor in 1996 and it took 2 acquisitions for our vendor to become Cogent.

Also, regarding OP's post - I don't call overselling plans when you specifying that it is not a dedicated port, but just unmetered port. But I think this is not what OP mean overselling, is that correct?

[tulix]

Originally Posted by USSHC_Colo

We don't oversell, and I'm very happy about that. That's how I'd like to continue in the future. We have excess capacity on the circuits we pay for and it works out just fine. As we put in additional fiber runs I'd like to keep it that way. I don't ever want to be in a position where we have more commit than capacity - It would only be a headache for everyone involved.

It's hard to gauge the market. WHT does have that feeling of looking for the lowest price, but there are definitely people with a lot of experience that know you frequently get what you pay for.

We had commits before we've placed on our network FCP. Since it took over control for our network and since we are optimizing for performance (not to save cost) we can't commit anymore to any vendor - whatever our FCP will send to them this is how much they will get paid. So why for us overselling almost impossible unless we sell shared services - but then we are specifying that in our plans.

[USSHC_Colo]

Originally Posted by tulix

I don't call overselling plans when you specifying that it is not a dedicated port, but just unmetered port. But I think this is not what OP mean overselling, is that correct?

If it's not a dedicated port for that client then I'd say that's not overselling. A shared 100 mb port would be just that, shared.

I know what you mean about not being close to $5/mbit. It's not terrible, I'm just not going to hit the same pricing that a carrier hotel would. I'm in an underground facility designed to be far away from metro areas... that just plain means there isn't cheap bandwidth flowing by me... I've got to lease or lay fiber for everything which bumps that price up. As we keep adding though those prices keep going down, so one of these days...

[hhw]

Overselling or not isn't the point. I doubt any network systematically calculates the sum capacities of all their customers/downstreams and ensure they have the capacity available. And even if they did, if all the customers happened to send traffic to a particular upstream or peer at a particular location, there can still easily be congestion.

The important factors for avoiding congestion are:
1) How much capacity is available on each and every link? Are there any links running hot?
2) How quickly can the network move traffic around in the event of unforeseen sudden and major increases in traffic? Do they run with enough capacity available on their other links at all times that this is even an option?

For this reason, I would say what you need from a network to ensure you won't see congestion issues is that they run all common links with a minimum 50% available, be they trunks, backbones, peers, or upstreams, or sufficient capacity for the sum amount of traffic resulting from multiple customers rapidly increasing their traffic simultaneously in a realistic worst-case scenario.

[iansltx]

Let me propose a different tack:

I, and the majority of Americans AFAIK, use coaxial cable to get to the internet. Capacity on that cable is Nx38 Mbps down and 10-30 Mbps up. The typical scenario is that a couple hundred homes will be on a given node, each with a 12-22 Mbps plan (with a few on 50 Mbps in my case). Thus the overselling ratio is 15:1 or higher. Yet nobody cares as long as everyone can max out their bandwidth when they want to. I'd put a term on this: responsible overselling.

Now I know that data center usage patterns are a bit different than residential broadband, and that overselling at that ratio would equal disaster unless you've got a HUGE pipe, preferably burstable. However the same principle applies: if your customers aren't adversely impacted by overselling, it's not at all a big deal.

FWIW, I'm working with Cogent on a few links (one DIA, one transport...bids are out on transport so they may not be the ultimate vendor) and they've told me a few times now that their core network is not oversold. Didn't say anything about their edge locations (which I am of course connected to) however they say they aren't oversold and I have no reason to believe otherwise. Peering link and edge congestion, maybe. But the core is clean :-)

[jweeb]

The price difference is marginal. If your websites are not too mission critical, it is alright if you are on a shared line. If you are lookig for quality and reliabilty, get something guaranteed.

[RyanD]

Originally Posted by tulix

I know what Ryan is talking about - WHT is a specific market, so why low end vendors, like Mzima, HE, Cogent are so popular in this market place. Top notch network is different for different markets - again, in our case we can't come even close to $5 per Mb/s, but we are talking about just different quality/markets.

While I would agree that HE and Cogent are not performance network vendors Mzima/PacketExchange on the other hand runs a very solid network. Our FCP's metrics actually place Mzima/PE as our most consistent performer. That may be due in large part to the fact that 95% of the traffic we see pass Mzima is immediately handed off to Level(3). We do not see them peering off alot of our traffic in Atlanta as we peer with most of the same entities they do on the AIX.

When running a performance network the Tier of a carrier is irrelevant. Your ultimate goal is to reach the eyeballs in the best path possible. Logically the larger your pool of paths the FCP has to choose from, the more optimal your route is going to be, this is why we run a diverse mixture of Tier 1's, Tier 2s, and private peering. Put it all in the mix and let our FCP send it where it needs to be.