95th Vs Per Gig
 

pros and cons for both

how much can you really save from either or?

95th sucks ass.....especially with my bandwidth graph

Pay for what you use, fair for you, fair for them.

if the top 5% is cut off how am i not saving money?

95th is terrible and costly thank god sinhost bills on average.

thanks brad

95th screws you on spikes. Candid bills on the avg outgoing only :thumbsup

Here's how 95% really works. Every x minutes (mostly 5) your
ISP is doing a measurement of your throughput. At the end of the
month all measurements are being ordered descending. The top
5% is taken off.....the FIRST measurement is your the level
you will get billed for.

So if you burst more than 5% on average per day (2.5 hours) then
you will get burnt on it. Let's assume on average you use 2Mbps
but between 10pm and 1am you burst up to 4Mbps average then
you will get billed for 4Mbps average usage per month.

So only if your traffic pattern is relatively flat......meaning no more
burst than 2.5 hours per day MAX it could be convenient.....but
most of the time it's convenient for your ISP.

DynaMite :2 cents:

Because that's just smoke and mirrors they're using to sell you that billing model.
Granted a lot of hosts don't have a choice, they're billed on 95th percentile from their carrier so that's how they have to bill you.

How are you not saving money?
Look at my biz as an example. I buy spots on The Hun, when I get a listing there I'll push upwards of 40mbps the whole first day.
I'll do that several times a month.

Last month I "averaged" 10.6 mbps, or about 3400 gigs.
If I was being billed on 95th percentile I'd have had to pay for 40mbps, but since I'm billed on average I paid for 10.6.

Make sense?

on AVG all the way!

lenny2 is a smart man.

95th only saves you money IF your traffic is consistently flat... since no one has a flat mrtg (unless they are maxxing out a capped line or something similar), 95th always costs you more.

Dont fall for the 5% free bs. One passwd hack, one instance of hotlinking sob's or a gallery listing pulling some good numbers can seriously fuck up your month.

To clarify, its the first measurement after the top 5% is taken off. And 5% per day is actually only 1.2 hours..........

Thanks that's what I meant first measurement after top 5% is
off and 5% is 1.2 hours....I got that mingled up as I got billed
according 90% in the past which does come down to 2.5 hours.

DynaMite :thumbsup

Its' all an illusion i see

Yeah, not too often dynamite is going to be wrong technically about somethin so i just had to jump on it.

Also look at that those who sell on 95% are most of the time 30-40% cheaper (if your not getting screwed) then those who sell on average. Alot of the time the two come out the same in the end.

That's a good point as well. I can buy 95th percentile bandwidth for half of what I'm paying now.....but in my case I'd still pay more.

If I was billed at 95th percentile I'd have to pay for 40mbps....so at 100 per meg (to use round numbers) that would be 4K.

On average I paid for 10.6mbps, so at 200 a meg (again round numbers) that's $2120. Almost half of what I'd pay at 95th percentile.

My advice to anyone would be to pay on average or per gig, because very rarely will 95th percentile actually save you money.

here ya go...it's alot but its accurate =)

Billing Methods Explained:

There are three common types of billing methods that IPPs generally use to bill their dedicated server customers; the "95th percentile", "bandwidth average", and "per gigabyte" based billing. All have their distinct advantages though we feel that the "95th percentile" method is the most balanced overall method.
Let's start with the "per gigabyte" based billing. Anyone familiar with "shared server" hosting is likely to be quite familiar with this type of billing. Quite simply, log files are kept of your web activity throughout the month and then at the end of the month you pay a specific charge for each gigabyte of data you transferred.

"per gigabyte" based billing usually includes both your incoming and outgoing data transfer -- so if you transfer 3000 gigabytes of outgoing transfer and 1000 gigabytes of incoming transfer then you are billed for a total of 4000 gigabytes of transfer.

The second common method of billing is "bandwidth capped". With this type of billing you select a bandwidth "cap" (usually in increments of 1 megabit per second) and you are restricted from transferring data beyond your "cap" amount. For instance, if you purchased a 20 megabit bandwidth cap then your web sites are free to transfer as much data as they need, up to 20 megabits per second. If you try to transfer beyond 20 megabits per second then your traffic is "slowed down" so that you cannot break the 20 megabit per second barrier.

"bandwidth average" based billing is usually for a "full duplex" network connection. This means that if you purchase a 20 megabit per second connection then you can transfer 20 megabits of incoming transfer and 20 megabits of outgoing transfer at the same time.

The "95th percentile" based billing is sort of a hybrid between the previous two billing methods. Throughout the month your incoming and outgoing bandwidth is graphed in five minute intervals. Every five minutes your transfer is totaled for the previous five minutes and an average transfer rate is derived. From that data a graph is created to show you your "real-time" bandwidth usage:

- Real-Time Bandwidth Graphs (updated every five minutes)

For example, if during the five minute interval you had 300 megabytes of outgoing transfer and 30 megabytes of incoming transfer your average outgoing transfer rate would be 8 megabits and your average incoming transfer rate would be 0.8 megabits. For reference, 8 megabits equals 1 Megabyte of data transferred every second (multiplied out for 300 seconds -- five minutes -- you get a total of 300 Megabytes of data transferred).

After the five minute samples are calculated the lowest of the two samples (usually the incoming data sample) is tossed out and the highest of the two samples (usually the outgoing data sample) is kept. This in effect gives you free incoming data transfer since the smaller of the data samples is tossed out completely.

At the end of 30 days we have 8,640 samples. At that time the top 5% of the samples (432 samples -- 36 hours worth) are tossed out. You are then billed on the highest remaining sample. This gives you 36 hours of transfer for free (as well as already having half of the samples tossed out). You get the "free" incoming bandwidth of "bandwidth valved" based billing as well as the burstability of "per gigabyte" billing. The best of both worlds.

"95th Percentile" versus "Bandwidth Valved"
The largest advantage of "bandwidth valved" billing is the predictable monthly costs. If you buy 20 megabits of bandwidth then you are charged for exactly 20 megabits of bandwidth.
There are, however, several drawbacks to "bandwidth valved" billing. The single largest drawback is the fact that you need to "buy to the peak". The traffic patterns for all web sites are filled with peaks and valleys. No web site in the world transfers a steady amount of bandwidth.

If, for instance, you have a web site that transfers around 8 megabits per second during the off-peak hours but bursts to around 20 megabits during the busy period of the day then you'll want to make sure you purchase 20 megabits if you're purchasing "valved" bandwidth. If you purchase less than 20 megabits -- say perhaps 15 megabits -- then your web sites will run noticeably slower during the peak times.

To make matter worse, when you purchase 15 megabits and your web sites are only doing 8 megabits in the off-peak times you are in effect paying for bandwidth that you are not using.

That leaves you in the situation where your web sites run slower during the peak times because you didn't purchase enough bandwidth and then you're paying for unused bandwidth in the off-peak times.

This is where the advantages of the 95th percentile billing shine. Because your bandwidth is not valved in any way, your web sites are free to burst to "full fast ethernet speed" during the peak times -- if it bursts to 20 or 25 megabits then that is what is sampled and recorded -- but during the off-peak times you'll only be paying for 8 megabits. You do not need to worry about having to "upgrade" your service just because you plan on increasing your usage later on in the month, and it takes the guess-work out of trying to figure out exactly how much bandwidth you really need.

Then, at the end of the month, your top 5% of the samples are tossed out. You get 36 hours of free transfer and you do not have to buy extra bandwidth just to sustain your peak times.

When it's all said and done, at the end of the month you may have had to purchase 20 megabits of "valved" bandwidth, but your 95th percentile may only be 13 megabits.

Combining Bandwidth Valving with 95th Percentile Billing
Looking for the predictability of valved bandwidth with the burstable connectivity of 95th percentile? No problem. All of our virtual servers have bandwidth valving capabilities built-in. If you have a web site that transfers between 10 and 20 megabits, for example, we can valve that virtual server at 20 megabits to insure that you never exceed the 20 megabit threshold. The billing will continue to be on the 95th percentile, so if at the end of the month your 95th percentile shows your bandwidth usage at 13 megabits you pay for only 13 megabits.
This gives you a virtual "ethernet pipe" and guarantees that your transfer will not exceed that amount -- just as with bandwidth valving -- along with the superior pricing model of the 95th percentile billing.


enjoy
:)

Out of curiosity, do you have a screen shot of your MRTG?
Ive never seen one that spikes to 40 a few times a month come anywhere near being charged on 40 mbps. Unless your spiking higher?

moose contact me ill splain it to you

All I know is one thing... my bandwidth bill dropped a lot when I switched from 95th percentile to average, so from my experience average is much better.

Yeah I can go take one....but like I said, if I do 5 hun listings a month, and each of them does 40mbps or so the first day (anywhere between 35-45 usually) then that's 120 hours of a 40mbps peak.

With 95th percentile billing you only get the top 36 hours knocked off, so I'd end up paying for damn near 40.

I'm gonna go eat dinner, I'll post a screen shot in a bit.

wow, great info. thank you

yep :glugglug

All methods are OK. I like average, but the host will charge you MORE per mbps for average. So if you're running regular traffic don't have any hacks or spikes that last for more than a day and a half a month, 95% will proabably be cheaper. Paying more for average is basically an insurance policy. Capped is a nice solution too. A nice cheap solution is to use 95%, but with a cap set to 1.5 - 2x what BW you plan to use. That way you save cash, and can't get fucked if a hacker gets your box.

first of all: excellent post, webair. very informative and it answered so many questions for many.

second: what hosts bill on average and what hosts bill on "valved bandwith"?

thats alot of info webair,

some many different ways to play the numbers

webair,
awesome response...having pushed pipe for evil akamai for three years i'm always amazed to see how many folks don't really understand bw and billing...excellent job. another point, 95/5 works awesome if you run international domains, also we put all our adult customers on mainstream tier 1 ISP's...why? The traffic nearly balances their 9-to-5 MRTG's perfectly...that is, it loads them up where they are weak (late hours and weekends)...we always insist on better pricing for this and frequently get it...we just signed a 100 Mbps deal at the 95th with a Tier 1 for $75/mbps and have an option to buy another 50 Mbps at $50 if the MRTG's prove to smooth out their graphs. Also, you can try offering them rolling billing over 90 day periods...helps to smooth any wierd bump from one month to the next...

just my .02,
R

hmm which Tier-1 provider did ya use rasta?? :glugglug

I disagree with the post that webair made earlier.
I've read those exact same words several times before on the faq page of companies who bill using the 95th percentile model.

Its written specifically to sell the customer on the 95th percentile model and I don't feel its very fair in its assesment of the different billing models.

There are tons of per gig hosts that don't charge for inbound and outbound traffic, only outbound. And you can be billed on average without having a capped line.

The host I'm with actually has prices for all 3 models on their site, you can pick what you want...per gig, on average, or 95th percentile.
Average most definitely works best for me, I don't see how you can go wrong paying for exactly what you use.

Here ya go Moose, screenshot of the MRTG.
This one isn't as drastic as most of them are, I was lazy last month so only 3 spikes. Usually there's more than that, so with 95th percentile I'd be paying a much much higher number than the average that I paid for.
http://www.sinxxxchange.com/links/mrtg.jpg

Guys,

Here is a great tool to calculate your bandwidth:
http://www.valkaryn.net/bwcalc/bwcalc.cgi

Enjoy:thumbsup

Not L3...but perhaps next time...that Sean Connery sure is a sexy spokesman...:Graucho

All hosts get charged 95th.

Well, almost. There are times when that is not true (such as cogent pipes, someone buying a GigE port and capping it at 300meg, or whatever). Unless the provider can afford to buy full GigE ports (not overly many.. most hosts this size choose (wisely) to instead of maxing out gige ports, to give more traffic to another provider for redunancy sake).

It's the industry standard for large bandwidth commits, and that's why it's passed down.

As a host, why would you want to take on a customer who say, spikes to 150mbit for 1 week, and then doesn't use anything the rest of the month? You'd get dinged for 150meg from your upstreams, and be left holding the bag.

Yes, your host COULD cap themselves, but then that causes problems. We prefer to just let customers use what their traffic demands rather than artificially ratelimiting you.

Now, average is better for the customer in most cases. This is EXACTLY the same as per-gig billing. (think about it). We offer both choices, but at the end of the day, both are going to be pretty close in price no matter what. No host worth their salt will price themselves to go out of business because they can't pay their upstream bills.

Now, as a hosts gets larger, it gets easier to "play the odds" at traffic patterns, and then you can start dropping the average billing a bit. The reason? Chances our one of your other 500 dedicated customers will be bursting around the same rate at a different time.. Basically using your huge aggregate usage to "even out" your usage graphs.

Ok.. I originally was just going to point out that per-GB hahahaha average per mbit, but I got carried away. :)

-Phil

LOL good catch man!

DynaMite :thumbsup

I agree with you here. ISP's are often trying to kiss you up for
only charging for outgoing (asymetric) traffic. In fact most of them
only pay for asymetric to their carrier aswell......OR they sell their
symetric traffic to other companies.

Hosting companies hardly have incoming traffic. Incoming traffic
just consists out of maintenance traffic and of course all the
requests from users to the servers but requests are only VERY
small packets. Just look at your MRTG stats and you will see
you have very little incoming traffic, upload a big file and you will
see it going up shortly.

When I was still in the ISP business we sold all incoming traffic
to a company that provided satelite internet links and to dialup/
DSL/cable guys. Most ISP's will have similar deals. They won't
just give it away and they certainly won't let it go to waste.

DynaMite :2 cents:

fun with mrtg graphs...

I have to go with lenny2 again here... that webair explanation is exactly the one used by hosts that cahrge 95th - there is a reason for that.

rarely do customers have a nice flat traffic attern to make 95th feasible.

Here's a few mrtg graphs:

client 1:

http://ezprovider.net/week1.png
http://ezprovider.net/month1.png

client 2:

http://ezprovider.net/week2.png
http://ezprovider.net/month2.png

client 3:

http://ezprovider.net/week3.png
http://ezprovider.net/month3.png


None of these clients would benefit from 95th... (ps - the cliets are japanese, thus the spikes at the early hours of the day)

Yes being billed on average is exactly the same as paying per gig. You pay for exactly what you use, nothing wrong with that.

There are hosts out there that don't get billed by their providers on 95th percentile, and are able to pass that benefit on to their customers.

:thumbsup :thumbsup :thumbsup

95th is the big shaving from hosters :321GFY

exactly...now all you have to do is choose the one thats best for you and talk your host into giving it to you...I hear alot of people saying that 95% is no good, that's not true at all, it can be a benefit for you if you do not spike alot =)

SHAMELESS PLUG: for the record webair will set you up anyway you like, your choice

:)

this is one of the most innaccurate statements I have ever heard =)

once again i am not defendiing any of these billing options as we offer our clients the choice to be capped, burstable or per gig...furthermore, whoever your hosting with should offer you the same options, in my opinion of course

per gig is the way to go... you always know where you stand

LOL, I always pay on Average :glugglug
When a hoster try to deal with me at 95th I say :321GFY

why not just buy the size pipe you need?