Lower plan or pay-per-resource?

The former, I'm pretty sure they've said they'll never go smaller as it's just not cost/resource effective.

That's what I thought.

I think a pay-per-resource plan could work but might be a little hard to setup and track, especially for CPU and memory usage.

It could also be a semi-fixed plan, like pay 10$/month for CPU/memory/disk and 0.05$/GB of transfer…

If a mere $20/month is causing you grief, then either your business model sucks, or you're just playing around and hosting for you is a plaything, not a necessity.

In either case, stop buying it if you can't afford it.

As to Linode coming up with a cheaper plan (no matter how you spin it) it probably can't happen. Anything less then $20/month and Linode probably loses money even if the the VPS isn't used at all after being setup.

Perhaps you've never had a business class, but there's this thing called "overhead". And no matter how small the actual amount of resources you use, Linode still has to manage the hardware, the hardware maintenance, the network, the network maintenance, the staff, the payroll, the insurance, the marketing, AND the billing - all of which means that at $20/month, they're already making squat (or close to it).

Yes, as I said, It's a personal server. I'm a student and sometimes a freelancer.

I can afford the $20/month. It's far cheaper than having a server at home running 24/7. That's not the problem. It just feel like I'm wasting resources: Money in my case, CPU/memory/disk/bandwidth in Linode's case.

I know that Linode has to pay everything you listed, but Amazon EC2 are billing the way I described (somewhat), and they have to pay the same bill than Linode.

See, I was thinking the same about my ISP. I have 100GB of bandwidth, but I almost never reach that amount. So I feel like I'm paying for something I don't use. Plus, they charge me when I bust. Why don't they charge me $X/GB?

With your home Internet connection, a good portion of your MRC goes towards things other than bandwidth: replacing modems, fixing physical plant, payroll for engineers and technicians, taxes, licensing, insurance, the whole deal. For the portion that goes towards bandwidth, the cost per GB exchanged with the Internet is almost certainly greater than that portion divided by 100 GB.

In other words, they're betting on the average user using much less than 100 GB, and the 100 GB limit is there to improve the odds.

Linode, I can only imagine, is a similar operation: I know I come nowhere near my monthly allotment. However, through the sacrifice of poor schmucks like me, the 200 GB is more generous than it otherwise would be.

Why not just ditch the complementary bandwidth, charge what it costs, and let me save a couple bucks while someone else pays a couple bucks more? Aside from the obvious collections, marketing, and customer retention problems, consider that large pipes aren't metered by gigabytes per month. They're measured by megabits per second. Linode probably pays their colocation providers $x/mo for some number of 1000Mb/sec pipes, plus perhaps $y/megabit/second to cover the peak traffic level transferred over each pipe (usually something like the 95th percentile five-minute average).

This means that Linode's bandwidth bill has no real relationship to how many gigabytes the customers move in a month. Per the Interesting Statistics, we can see that Linode averages 2100 Mb/sec for (5000 cores / 8 cores per server * 40 standard Linode units per server * 200 GB per month per standard Linode unit) = 5000 TB/mo of allocated customer bandwidth. If I'm doing my math right, if everyone ran through their entire 200 GB, that's an average of 15400 Mb/sec.

(Note: I did use 1000 instead of 1024, to keep the math easy.)

It stands to reason, therefore, that people don't go through their entire bandwidth allotment. The number's probably closer to 27 GB per standard Linode unit. If we wild-assedly-guess that 25% of the marginal cost goes towards bandwidth (that's $5 per standard Linode unit), the out-of-pocket cost per GB should be somewhere around $0.20/GB.

In other words, you'd probably be looking at paying $15/mo plus $0.20/GB. If you use 50 GB, you're paying $25/mo for what you now pay $20/mo for.

And as it turns out, I did all these calculations and all this work and I could have just looked at another major VPS provider's price list (you know, the ones who charge for actual bandwidth used) and used their numbers. Shit.

@subb:

I know that Linode has to pay everything you listed, but Amazon EC2 are billing the way I described (somewhat), and they have to pay the same bill than Linode.

EC2 is a completely different pricing model. Linode has chosen this model. If you want EC2's model, go to EC2.

@subb:

I know that Linode has to pay everything you listed, but Amazon EC2 are billing the way I described (somewhat), and they have to pay the same bill than Linode.
Have you estimated actual costs for EC2 with your usage pattern? In my experience it doesn't take much to exceed Linode's pricing.

For example, a small EC2 Linux instance, with 10GB of persistent EBS storage looks like it would cross the $19.95/month point if it is active for 222 hours that month, or only about 30% of the time. That's a pretty low cross-over point. And it assumes no charging for network transfers, which is indicated to go away at the end of this month. So yes, you can get resource-based billing with Amazon, but there's actually an overhead to that built into the pricing structure.

$20/mo for a 360 can certainly add up, especially with multiple Linodes, so desiring a lower price point is understandable. But it's a pretty darn good price point (<3 cents/hour) for the resources it makes available. I'm not sure how much lower it could realistically go and maintain the overall features that I find so desirable with Linode.

Even if pro-rated against actual usage, I think it comes out favorably with other ways of delivering a similar service, especially when taking the DC and networking benefits into account. I'm actually amazed at how low it is.

– David

Thanks hoopycat for your super complete anwser. I had a feeling when I asked the original question that it would end up costing more most of the time, but I had nothing to base it on. That's why I asked the question in the first place.

That is what happens when I browse the forums while taking a break from math homework. :-)

There is these guys for "pay for what you use" service: https://www.nearlyfreespeech.net/

The cheapest plan on EC2 is like $80/month, and it often performs worse than a Linode 360. That's what you get when you try to go pay-per-resource through and through. The same is true of nearlyfreespeech: the performance sucks for anything dynamic.

In fact, you could actually use a Linode like you'd use an EC2 instance: fire one up when you need it, get rid of it when you don't need it, and fire it back on later. Everything is pro-rated to the day (unlike EC2 where it's per-hour), and it's a lot cheaper than EC2.

Spend some time with it, you probably get even create a StackScript to restore images/data saved in S3 or something.

The world is turning into one big Walmart shopper, ignoring quality, ignoring service, ignoring support, and just mindlessly droning - cheaper, cheaper, cheaper.

Linode - don't change a thing - you're doing great just how you are.

@hybinet:

Everything is pro-rated to the day (unlike EC2 where it's per-hour), and it's a lot cheaper than EC2.

linode appears to be prorated to much less than a day. Last one I built, I ended up creating and deleting a few times and the cost was for the prorated month kept going down a couple cents each time.

I personally don't like being nickel and dimed to death.

Most months I use <10GB of bandwidth. Others, I do crazy things like seed massive torrents @ 40mbit for about 30 minutes. And… I use a lot more than 10GB.

You aren't paying for resources that are "wasted" you're paying to reserve their use.

http://en.wiktionary.org/wiki/tomato_tomato

Linode, you are fantastic. Please don't change a thing. I have tried EC2 and Linode, and much prefer Linode. The performance is better, the downtimes that are few and far between are explained, I get quick support via email and I can usually chat to the australian dude in IRC. EC2 was expensive with usage charged separately, and had mysterious outages that couldn't be explained.

Keep up the good work Linode.

@subb:

I can afford the $20/month. It's far cheaper than having a server at home running 24/7. That's not the problem. It just feel like I'm wasting resources: Money in my case, CPU/memory/disk/bandwidth in Linode's case.

If you feel like you're wasting your linode install something useful on it, like a TOR router or a scientific computing client for the World Community Grid.

@khedron:

a scientific computing client for the World Community Grid.

NO! Don't suggest stuff like this, it's completely inappropriate on a shared environment

I host websites for $5 to family and friends. I almost break even ($30) on my 720 plan. I can't "play", but I don't want to, I want it up and running all the time. I play in a VM at home.

@subb:

I was wondering if you were planning to create a plan with lower specs (bandwidth and disk space in particular) or, even better, a pay-per-resource-used plan.
If you want a cheap-and-cheerful VPN to mess about with, you could have a look at http://www.lowendbox.com/. Keep the Linode for the serious stuff, and have a dirt-cheap box for experiments.

Wow. This thread went a little farther than I expected. I was not asking for a change, I was just wondering why Linode's business model is the way it is…

@cager Thanks for the useful link.

@glg Why is that?

@subb:

@glg Why is that?

Because world computing grid applications are designed to use idle disk cpu time/ram etc, when you're on a linode vps you're sharing resources with X number of other nodes so if you're running one you're using resources other nodes can be using.

Yea, but isn't that the point? Using the resources I'm paying for?

Or did you mean the WCG application try to use the whole (physical) CPU/RAM/Disk?

I dont think you quite understand the shared part of the cpu..

Think of it like this..

You have two cpu's and two users.

if User 1 is doing nothing, user 2 can use both, but if they both want to do stuff then they only get 1 cpu each.

Now if User 1 puts on something like seti@home or any of those thing they are always going to be using the CPU.. so your not sharing or allowing other people to use it. So rather then user 2 having the potential to use 2 cpu's at most, they can only use 1 now..

Does that make more sense?

@subb:

Yea, but isn't that the point? Using the resources I'm paying for?
We actually all benefit from the statistical sharing of resources (at least in terms of CPU and I/O - memory is guaranteed and fixed) beyond what you'd get if you equally divided up the host. Essentially you can get more than you pay for, but only when others aren't using it.

On a Linode 360 (oops, 512) there are roughly 40 Linodes on a single 8-processor host. Assuming a 2.5GHz processor (fairly typical host), that means that if everyone were using as much CPU as possible, we'd all get about 0.5GHz - probably closer to 0.4GHz if you take host usage into account. Nothing wrong with that per-se - it's about all we are guaranteed, but it'd be much slower than typical behavior.

In reality, we all tend to get far more than that (and can burst to a full 4 processors) because often many other nodes are idle at any given instant in time, and Linode doesn't limit us strictly to the portion of CPU that should be allocated to each Linode if evenly divided.

I/O is shared similarly in that there is some overall bandwidth available to the disk subsystem, and at any given moment it will be shared amongst all the Linode's performing I/O. So each Linode gets higher performance since on average many of the others are not doing I/O at the exact same instant.

That's still not to say that if you need the CPU or I/O you shouldn't just use it. It will be evenly shared amongst everyone on the same host who wants it at the same time, and that's what we all bought into when we selected our plan.

But running a completely CPU or I/O bound application constantly when not actually needed for your Linode's operation isn't quite "neighborly," as it will create contention that need not exist, and lower the processing power or I/O available to others. Heavy I/O is actually even worse than CPU as it's generally a more constrained resource.

We're all sort of in the same boat and have vested interests in using what we need, but not burning resources unnecessarily.

– David

@db3l Thanks.

I guess I can conclude with this : If you're one of my linode's neighbors, enjoy! :wink: