Cloud Price Wars – What about the network?


The cloud price wars that began at the end of March have been all about compute and storage pricing. I don’t recall hearing network pricing being mentioned at all; and indeed there haven’t been any major shifts in network pricing.

Photo credit: Datacenter World

Network is perhaps now the largest hidden cost of using major IaaS providers, and also one of their highest margin services.

Let’s take a practical (and personal) example. At the start of last year the Raspberry Pi images for OpenELEC that I was hosting on my Pi Chimney site were being downloaded around 35,000 times a month generating 3.5TB of network traffic.

Let’s have a look at my options using present prices:

  • AWS EC2 assuming I could use a free tier instance and that it wouldn’t melt under the load then my data transfer out of Amazon EC2 to Internet would be $0.12 per GB for a total of $429.96
    • I could have put the files into S3 and paid the same transfer cost of $429.96 (but not worried about server load)
    • CloudFront wouldn’t make any difference either, as that’s also $0.12/GB for a total of $429.96, though maybe the service would be faster
  • GCE pricing is pretty close to AWS. The first TB is $0.12, and the 1-10TB is a little cheaper at $0.11 for a total of $404.48 – a paltry saving of $25.48
  • Azure is also $.12 for the first 10TB, though it offers 5GB free where Amazon only offers 1GB (and Google offers none), so it’s fractionally cheaper at $429.48
  • The first offer I find on LowEndBox (a marketplace for cheap virtual private servers) with a 4TB bandwidth allowance is XenPower’s XP-XL at $6.68 – a saving of almost $400 against the cheapest major IaaS vendor!

When this was a real issue for me I used a VPS provider that included 3TB of network with the $10/month server I was using, I would switch traffic at the end of the month to a second VPS I had that included 1TB of network in the $6.99/month. The site has subsequently moved to GreenQloud where they provide complimentary hosting in exchange for me advertising their sponsorship[1].

High margin

Let’s for the sake of argument say that the $6.68 for a VPS is just bandwidth – the compute, storage and IP address come for free (which they don’t). Furthermore let’s say that’s cost price for the VPS provider. In that case we can take an Amazon charge of $480 (for 4000GB) and calculate a gross margin of 98.6%.
Of course VPS providers don’t get compute and storage and IPs for free[2], so it’s reasonable to assume that Amazon (and it’s competitors) are getting >99% gross margin on bandwidth. Nice business if you can get it.

Higher margin

Of course the calculation above assumes that the VPS provider and Amazon pay the same for their bandwidth, which of course they don’t. Amazon has a far better bargaining position with its telecoms providers than XenPower (or any other VPS provider likely to show up on LowEndBox). It’s fair to assume that Amazon pays much less for its bandwidth than XenPower does.

What is the cost of bandwidth anyway?

Bandwidth within co-location centres gets charged by the MBps. A typical rate at present is $2/month for a 100Mbps link, or $1.50/month for 1Gbps. 10Gbps links are cheaper still, but nobody seems to want to publish pricing information on the Internet. Driving one of those 1Gbps links at full speed would translate into a cost of $0.00474 per GB/month. That’s still a lot more than the $0.00167 per GB/month from XenPower.

Tiered pricing

My example fits into the 0-10TB price band for the big clouds. Amazon (and its competitors) have a tiered pricing structure where the cost drops to $0.05/GB for 150-500TB, and there are three bands beyond that with a price of ‘contact us’. It’s not at all clear why there’s consistent pricing for instances but ramped pricing for network. What is clear is that the cloud providers can sell transfer in bulk at less than $0.05/GB.

Bandwidth != transfer

Bandwidth is the capacity of a network link, whist transfer is the actual traffic pushed across it. In real life network utilization is bursty, and so it’s necessary to buy more bandwidth than the notional transfer being supported over time (otherwise the network becomes a bottleneck).

So how do the VPS providers do it?

It’s a mixture of factors:

  • Throttling – the VPS provider will have an Internet connection with a given bandwidth, and once that’s saturated then customers will start to notice additional latency. Individual servers might also be throttled.
  • Capping – once a bandwidth allocation is used then the VPS might be switched off until the end of the billing cycle.
  • Mostly unused – the VPS provider can build a statistical model of how much transfer actually gets used, which won’t always be the max for every customer.

All of this adds up to VPS providers paying for a limited amount of bandwidth.

Photo credit: Rusted Reality post

So what’s different about IaaS?

IaaS providers are charging by transfer actually used without imposing caps, and having to buy sufficient bandwidth to support that transfer without throttling, which involves some degree of oversupply from a capacity management perspective. That said, it still looks like a very high margin part of their business, and one that hasn’t become cheaper as a result of the cloud price wars.


[1] GreenQloud offers 1TB free with its servers then charges $0.08 per GB for transfer, so one of their Nano servers would come in at $212.10/month for my 3.5TB workload.
[2] IPv4 addresses are starting to become a significant cost contributor to very low end VPS offerings, which is why IPv6 only packages are starting to crop up.

This post originally appeared on the CohesiveFT blog as part of the Cloud Price Wars Series

5 Responses to “Cloud Price Wars – What about the network?”

  1. 1 @ndy

    If you buy a 100Mbps (or 1 or 10G) link in a colo facility you won’t necessarily get any transfer included. Are the costs you cite ($2/m) just the port fee?

    For “low” bandwidth use you can pay per GB, TB or 10s of TB.

    However, bandwidth is usually charged for at a “commit” level, independent of the underlying link: you pay for the *speed* that you use. This is usually calculated at the 95-percentile of the maximum *speed* you pushed that month. Users “commit” in advance and pay that price even if they don’t use it. If they go over then the charges are usually much higher. A typical “commit” on a 100Mbps link might be only a few Mbps. A quick google suggests ( that you might be paying £3 per Mbps per month so 10Mbps would cost you £30 and allow you to transfer 10Mbps constantly for 95% of the time (and bursting up to the link capacity for the remainder of the time).

    Let’s say there are 2,628,000 seconds in a month. Therefore you can transfer about 2.4TByte if you push 10Mbps for 95% of the time and nothing otherwise. That’s about £0.012 / GByte (excluding the port fee).

    Now, Amazon with have a much higher commit than that and will have many links. Some of their links will be “peering” links where the costs involved are different: commit usually applies to “transit”. “transit” is a connection where the provider will route your traffic to anywhere on The Internet. A “peering” connection is a link to just one other network (such as Google, or BBC, for example). The bandwidth is usually free across these “peering” links and the fees are just the cost of infrastructure; port fees, cabling, etc. (And “syndication” / “membership” and other administrative fees in the case of exchanges such as LINX.)

    Amazon (and other big network operators) will use peering to strategically manage their bandwidth bills and send as little data as possible via transit.

    The actual price the customer pays for bandwidth is a price based on the cost of the underlying asset, aggregated over all their customers, as well as Amazon’s costs involved in the equipment, peering arrangements, staff and management.

    At $0.12/GB you’re “only” paying Amazon a ~5x ($2 -> £1) for what you could possibly arrange for yourself (at cost), provided your had a really stable traffic profile.

  2. Probably also worth mentioning that Softlayer offer free inbound and internal private networking (across all their DCs), and their servers generally come with multi-TBs of outbound traffic included in the price. This remains one of their big advantages and is easy to miss when you’re comparing pricing between the big 3 public cloud providers (Amazon, Google and Microsoft).

    • Most of the clouds only charge for outbound Internet traffic (though some also charge for traffic between regions as it’s not just on their LAN then). I do agree that it’s worth looking out for bundles that include outbound traffic – the whole point of this post was to draw attention to what can be a hidden cost (and also a high margin item for the service providers).

  1. 1 Internets of Interest - 27 July 2014 - EtherealMind
  2. 2 Sysadmin Sunday 189 - Server Density Blog

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: