Results 1 to 29 of 29
  1. #1

    Considering Co-location, please help.

    I am considering co-location vs. dedicated server for my next project. I am very confused as to the price difference. I've found some reliable providers of dedicated servers that will give me a nice p4 dual xeon for $250 + 1000 gigs bandwidth.

    However, it seems like colocation should be cheaper. But when looking at co-lo I see very high bandwidth prices and in the end it seems like I would be looking at a much higher price tag. 1 Mbps seems very slow to me, and while I don't plan on using more that 500 gigs, I would like it to be able to tranfer at 100Mbs if needed. Are there any co-location providers that offer this kind of bandwidth setup?

    Finally, do most co-location dudes fix your server quickly if you give them spare parts to keep around?


    Thanks.

  2. #2
    Join Date
    Aug 2002
    Location
    Atlanta, GA
    Posts
    1,114
    The reason for the price difference is most hosts know many dedicated clients won't use all of their bandwidth while many colocations clients will. Most hosting companies are willing to oversell dedicated bandwidth but not colocation.

    If you want the ability to push more bandiwdth you need to have a host with burstable bandwidth something like 10/100 burtsable. You may also want to look at one that has the 95% usage billing method to reduce costs on bandwidth use that has some higher peaks.

    Most quality colocation hosts will store spares for you and do parts replacement for free or for a nominal fee.
    SiteSouth
    Atlanta, GA and Las Vegas, NV. Colocation

  3. #3
    Join Date
    May 2003
    Posts
    472
    It would probably be best for you to get a dedicated server.

    Almost all colocation providers now charge a small fee for rack space and then a fee per mbps. To be able to burst to 100mbps would cost you a fortune.

  4. #4
    Join Date
    Aug 2002
    Location
    Atlanta, GA
    Posts
    1,114
    The ability to burst your speed doesn't cost any more. It's the bandwidth use that costs more. It cost the colo company the same for everything it's just the actual bandwidth use.

    As an example, we provide 1 Mbs, 10 Mbs or 100 Mbs ports. It cost the same per Mbs for any of these access ports but your billed on the actual bandwidth used not the speed of the port. Doesn't matter if it's a 1, 10 or 100 Mbs port.
    SiteSouth
    Atlanta, GA and Las Vegas, NV. Colocation

  5. #5
    Bursting to a 100mbps is a bit ambitious in my opinion, I have seen very few servers that are capable of doing that. Once you get anywhere near that speed so many things can slow it.

  6. #6
    Join Date
    Jul 2002
    Location
    San Luis Obispo, CA
    Posts
    818

    Re: Considering Co-location, please help.

    1mbps = 324GB, usually colocation providers give you a 10/100 fully burstable port

    Originally posted by krisr
    I am considering co-location vs. dedicated server for my next project. I am very confused as to the price difference. I've found some reliable providers of dedicated servers that will give me a nice p4 dual xeon for $250 + 1000 gigs bandwidth.

    However, it seems like colocation should be cheaper. But when looking at co-lo I see very high bandwidth prices and in the end it seems like I would be looking at a much higher price tag. 1 Mbps seems very slow to me, and while I don't plan on using more that 500 gigs, I would like it to be able to tranfer at 100Mbs if needed. Are there any co-location providers that offer this kind of bandwidth setup?

    Finally, do most co-location dudes fix your server quickly if you give them spare parts to keep around?


    Thanks.
    Nick Twaddell
    WebSpace Solutions - Custom E-Solutions
    Fast, Reliable, Affordable Web Hosting

  7. #7
    Join Date
    Nov 2002
    Posts
    2,780
    100Mbps isn't that good an idea. I have to agree on backup and restore they rock, but when a worm is sneaking through your network, you'll see a 100Mbps port to be a big trouble. In our experience, 100Mbps port + Windows = Disater.

  8. #8
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083

    Re: Re: Considering Co-location, please help.

    Originally posted by ntwaddel
    1mbps = 324GB, usually colocation providers give you a 10/100 fully burstable port
    I'd be careful about throwing that statement around in an environment where people might actually believe it's true of either the dedicated or co-location market, and stake money on it.

    If you are billed by average utilization, then in a 30-day month you will indeed get 324GB out of 1Mb/sec mean utilization.

    However, most providers of dedicated servers and co-location bill this type of customer based on a 95th-percentile burstable utilization figure, which is far from the mean utilization. It's not unfair, as most ISPs pay for their bandwidth via this same model.

    What is 95th-percentile? To determine your bandwidth usage on dedicated and co-location, your host polls or samples your ethernet port at regular intervals, say every 60 - 600 seconds. They record the number of bytes transmitted and received since the last sample. This number is then converted from a time-frame and absolute quantity to a rate, for example converting from 70 miles in two hours to 35 miles per hour.

    In a 95th percentile model, the samples for each time-frame (60 - 600 seconds) throughout the billing period (month) are sorted, and the highest 5% of the samples are thrown out. The highest remaining sample is used as your billable figure.

    What does this mean in the real world? Average hosting traffic ends up at a meer 59% efficiency, with about 190GBytes transferred for every Mbit/sec billed in a 95th-percentile model. You don't have to take my word for it, though; figure it out based on your own traffic, because that's what matters to you.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  9. #9
    Join Date
    Mar 2004
    Location
    Corning, NY
    Posts
    12
    The 95th percentile billing does not affect how much data is transferred, but what you pay. It is, therefore, correct in assuming that the amount of data you actually transfer in a month will actually exceed your 95th percentile figure. THat is to say if your 95th percentile number is 1Mbps, then you actually have transferred more than 324GB for the month.

  10. #10
    Join Date
    Aug 2002
    Location
    Atlanta, GA
    Posts
    1,114
    Just look here for an explanation:

    http://www.sitesouth.com/monthly.htm

    SiteSouth
    Atlanta, GA and Las Vegas, NV. Colocation

  11. #11
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by hoangelos
    The 95th percentile billing does not affect how much data is transferred, but what you pay. It is, therefore, correct in assuming that the amount of data you actually transfer in a month will actually exceed your 95th percentile figure. THat is to say if your 95th percentile number is 1Mbps, then you actually have transferred more than 324GB for the month.
    No, that is not true. Please re-read my earlier post and tell me what you don't understand, so I can attempt to clarify for you. I assure you, I have many customers for which both figures are calculated for internal use. I am not posting a baseless assumption as if it were fact, as you are.

    If your average utilization for a 30 day month were 1Mb/s, then you have transferred 324GB during that month. The same is most certainly not true for services billed on a 95th-percentile model.
    Last edited by jsw6; 03-28-2004 at 01:01 PM.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  12. #12
    Join Date
    Sep 2002
    Location
    Orange County, CA
    Posts
    383
    It is important to note that not all ISPs calculate or gather samples for 95/5 in the same manner. I have seen some very "creative" methods of calculation.

    You should always ask to be shown exactly how your provider gathers their data and what method they use to calculate the samples taken.

    Also, there are a few providers that I've seen that bill 97/3. Hurricane Electric (he.net) did (does?) this.

  13. #13
    Join Date
    Mar 2004
    Location
    Corning, NY
    Posts
    12
    Originally posted by jsw6
    No, that is not true. Please re-read my earlier post and tell me what you don't understand, so I can attempt to clarify for you. I assure you, I have many customers for which both figures are calculated for internal use. I am not posting a baseless assumption as if it were fact, as you are.

    If your average utilization for a 30 day month were 1Mb/s, then you have transferred 324GB during that month. The same is most certainly not true for services billed on a 95th-percentile model.
    Look at the example that was given: http://www.sitesouth.com/monthly.htm.

    As you can see this person is billed for the transfer of around 107 Gb. I am saying that the person REALLY transferred more than this. They may not be billed for the rest, but the rest was indeed transferred. Am I missing anything. My assumption is that the average (not the 95th percentile) is the actual amount of data transferred.

  14. #14

    Question

    Ok, so this is what I am confused about. 1Mbs is 1 million bits per second which is only 128Kilo bytes per second. So if a provider is billing you based on a mega bit per second cap, what happens if someone downloads a file off of your server with a t3 line? Even a dsl line would break that speed. I must not be understanding something here. Thanks.

  15. #15
    Join Date
    Aug 2000
    Location
    Sheffield, South Yorks
    Posts
    3,480
    We record Actual usage (In bits transfered), Average Transfer and 95th Percentile for ALL customers, and I can say that 1Mbit/s on 95th never correlates to 324GB in any month, more like <200GB, for some customers it's a little more, maybe 210GB in a month, but not often. So saying 1Mbit/s on 95th% means you've transfered more than 324GB isn't going to be correct in 99% of cases.
    Karl Austin :: KDA Web Services Ltd.
    UK Business Hosting and Managed Servers - Hosting for Business Users :: 0800 5429 764
    Call us today and ask about our hosting solutions.

  16. #16
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by hoangelos
    Look at the example that was given: http://www.sitesouth.com/monthly.htm.

    As you can see this person is billed for the transfer of around 107 Gb. I am saying that the person REALLY transferred more than this. They may not be billed for the rest, but the rest was indeed transferred. Am I missing anything. My assumption is that the average (not the 95th percentile) is the actual amount of data transferred.
    I've visited the link you posted, and I'm sorry to say that you just don't understand the information they are presenting. If you'd care to pull up the example URL above again, and read my walk-through below, I think we can clear this up.

    They report the 95th-percentile utilization at 337Kbit/sec, which looks reasonable based on the graph (although you need the numeric samples to be sure; it's tough to eyeball!) They further report that the amount you would be billed for would be 107GB/mo, which is, unsurprisingly, 107/324 == 0.33Mbit/sec. You've gotten this far on your own.

    You'll also notice an average out indication on their graph, which they report to be 196Kbit/sec. This is where we've got some confusion to work out! 196Kbit/sec is the total volume of data actually transferred over the course of the billing period! If you've got a calculator at the ready, you can find that 0.196*324 == 63.504GB/mo. The example customer in that URL really transferred only 63.5GB, but was billed at 0.33Mbit/sec; or 107GB/mo rate due to the way the popular 95th-percentile billing model works.

    If you like arithmetic, as I do, you will further calculate that 63.5 / 107 == 59.34%; meaning that although the customer paid for a 107GB traffic rate, they actually utilized only 59.34% of that. They paid for 141Kbit/sec or 43.5GB/mo more bandwidth than they actually used! They didn't actually save money on 95th-percentile, as you suggest; in fact they paid for 68.5% more than they used.

    If you scroll up a bit in this thread, you will find and earlier post from myself, indicating that 59% efficiency is typical of web hosting customers on a 95th-percentile billing model, and that is almost exactly what is shown in the example at the above URL. This is not coincidence -- I based my statement on my own billing data for many customers.

    Originally posted by KDAWebServices
    We record Actual usage (In bits transfered), Average Transfer and 95th Percentile for ALL customers, and I can say that 1Mbit/s on 95th never correlates to 324GB in any month, more like <200GB, for some customers it's a little more, maybe 210GB in a month, but not often. So saying 1Mbit/s on 95th% means you've transfered more than 324GB isn't going to be correct in 99% of cases.
    Kind thanks for posting your figures as well. It's nice to see more folks with actual data to back up their position posting on a complex subject like this. I believe it's important that all webmasters have a firm understanding of the model under which they are billed. 95th-percentile shouldn't be an industry black box, and when folks take the time to explain it thoroughly it not only helps webmasters avoid surprising invoices; but keeps hosting companies from being put in the unfortunate position of having a customer who believes they are being over-charged!
    Last edited by jsw6; 03-28-2004 at 03:49 PM.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  17. #17
    Sorry to interrupt, but I am still curious as to how this all works out in the end. Here is my second post with more detail:

    1Mbs is 1 million bits per second which is only 128Kilo bytes per second. So if a provider is billing you based on a mega bit per second cap, what happens if someone downloads a file off of your server with a t3 line? Even a dsl line would break that speed. I must not be understanding something here. So if the majority of my customers are using dsl lines then I am going to be using an average of 1Mbs? And if there are consistently 5 users connected i will be using even more than that. How does this all factor in???? I realize the majority of page hits are going to be less than 80k but what if I serve files as well?

    Thanks.

  18. #18
    Join Date
    Aug 2002
    Location
    Atlanta, GA
    Posts
    1,114
    196Kbit/sec is the total volume of data actually transferred over the course of the billing period!
    Read the chart. 196Kbit is the "average" volume not the "total" volume.

    The math we provide is the same math that 90% of the bandwidth providers use.

    95% is best for clients who have large, short terms bursts of bandwidth. As an example: we have a client that sits at about 300 Gbs for the month. However, one day a week, every week, he hits 30 Mbs for about an hour. He's basically billed for the 300 Gb per month even though he's hitting 30 Mbs for 3-4 hours per month. In this case it is to his advantage because the 95% billing takes those peaks out.

    This debate comes up every three months or so. There are advantages, for both the customer and the host, to use either of these two methods of billing. As a client you should choose the billing method that fits your needs best.
    SiteSouth
    Atlanta, GA and Las Vegas, NV. Colocation

  19. #19
    I understand the math. I guess what I am asking about is what a large site (say a shared hosting service with 100 standard clients) would use in terms of average bandwidth. Everyone I know these days has at least 1.5Mbs downstream via dsl/cable. However, I don't really know what that would equate to in throughput when browsing say an auction site. I know tcp starts slow, so I suspect you generally would not break the 1Mbs limit if you only have a few simulatenous users and no large downloads. If anyone could give me an idea of average bandwidth requirements for a medium sized shared hosting business that would help. This is not my business but its close enough.

    Thanks again.

  20. #20
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by mgphoto
    Read the chart. 196Kbit is the "average" volume not the "total" volume.
    mgphoto, you simply do not understand the math here! If 196Kbit/sec is the average transfer rate over the entire month, and that is how it is labelled on your example and what you have just stated above, then there is no way you can twist words or math to come up with anything other than 196 * 2,592,000 / 8,000,000 == 63.504. The units are 196Kbit/second, 2,592,000 seconds in each 30 day calendar month, and 8,000,000 kilobits per gigabyte.

    The result is 63.504 gigabytes transferred, which, as it so happens, is the same figure you get if you take 0.196*324; or 196Kbit/sec times 324 GBytes per Mbit/s sustained per month.

    Let's put it another way. If I drive my car at an average of 4 miles-per-hour for a month, and there are 720 hours in a month, then I drive 2,880 miles that month. These are not numbers you can refute, a mean average is a mean average.

    If you can't understand the simple arithmetic I've demonstrated above, then feel free to telephone me and I'll walk you through it once again with members of your staff if you so choose. You misunderstand the billing model, and your incorrect statements about it are going to cause other folks to believe what you do as well.

    95% is best for clients who have large, short terms bursts of bandwidth. As an example: we have a client that sits at about 300 Gbs for the month. However, one day a week, every week, he hits 30 Mbs for about an hour. He's basically billed for the 300 Gb per month even though he's hitting 30 Mbs for 3-4 hours per month. In this case it is to his advantage because the 95% billing takes those peaks out.
    95th-percentile is not, in fact, best for the huge majority of web sites and hosts. You just don't understand the math, and you've referenced an example scenario which I think everyone can agree is both extreme and a-typical of hosting traffic; and at the same time, you've failed to post enough detail so anyone can tell you whether or not the customer would be better off being billed at a flat-rate per GByte!

    While I don't expect you to have access to the raw samples, I would love to see an MRTG / RRD graph including mean (avg) utilization for the above billing period. By my calculations, your customer's 30Mb/s bursts for 4 hours each month make up only 54GBytes of traffic -- which will still not offset the typical 42% loss in the 95th-percentile model. If they were a typical customer, without those huge bursts, I would expect their mean port utilization to be approximately 0.537Mbit/sec, even though the 300GB you say they are billed for puts their invoice at 0.926Mbit/sec; a difference of 125.7GBytes/mo! There need to be some more good customer bursts on that graph to make up the difference between 125.7GB and 54GB transferred in your 4 one hour windows.

    While the extreme example you have used may be one of the rare cases where 95th-percentile is indeed an advantage to the customer, I doubt it; and I would love the oppertunity to examine more detailed data so we can find out for sure while this thread is fresh.

    Why do you think your own company's example precisely matches my predicted actual to billed traffic efficiency ratio? You can twist words and you can come up with cinderella examples with huge bursts, but your own example agrees with me almost exactly. The numbers do not lie.

    This debate comes up every three months or so. There are advantages, for both the customer and the host, to use either of these two methods of billing. As a client you should choose the billing method that fits your needs best.
    It's very hard to make an argument in favor of 95th-percentile, unless a host will not offer you bursting capabilities on any other billing model (which is common.) The burstable bandwidth model is expensive for transit providers to operate, and that expense is passed on to hosts and customers. The model doesn't exist to make it potentially cheaper to transfer data, it was invented to allow the tier-1 ISPs to recover the cost of over-building their networks in order to carry the burst traffic.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  21. #21
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by krisr
    If anyone could give me an idea of average bandwidth requirements for a medium sized shared hosting business that would help. This is not my business but its close enough.
    Obviously your bandwidth requirements will depend on your customer traffic. You can bank on about 190GB of sellable transfers per 1Mb/s of bandwidth you must buy from your provider(s). Thus, if your bandwidth cost were $80/Mbit, you would be looking at a cost of 80 / 190 or about $0.421/GByte. Again, this is with typical hosting traffic patterns, and your milage may vary.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  22. #22
    Join Date
    Aug 2002
    Location
    Atlanta, GA
    Posts
    1,114
    Sorry Jeff but not only do I disagree with you but the sample on the site is based upon upon Tobias Oetiker's work. Tobias just happens to be the guy that wrote the initial MRTG software and runs the MRTG site. You will see samples of the same chart on the MRTG site.

    The math and process is confirmed by Level3 and ATT. I guess Level3 and ATT just can't figure this whole bandwidth thing out.
    SiteSouth
    Atlanta, GA and Las Vegas, NV. Colocation

  23. #23
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083

    *

    Originally posted by mgphoto
    Sorry Jeff but not only do I disagree with you but the sample on the site is based upon upon Tobias Oetiker's work. Tobias just happens to be the guy that wrote the initial MRTG software and runs the MRTG site. You will see samples of the same chart on the MRTG site.

    The math and process is confirmed by Level3 and ATT. I guess Level3 and ATT just can't figure this whole bandwidth thing out.
    Post more details as I requested in my previous reply.. You still can't demonstrate why an average utilization rate of 196Kb/s during a month is somehow not equal to 63.5GBytes transferred; nor can you refute any of my other, mathematically backed, statements. I post my telephone number in my .sig; feel free to call me, or post yours and we will discuss this further.

    It just happens that you don't understand how to interpret the sample chart. And indeed, I have contributed code to RRDtool, which, as you probably know, is the successor to MRTG also maintained by Oetiker.

    Get in touch with me so we can settle this off the boards, as you are just making yourself appear foolish by calling me wrong without any math to back it up; and I am starting to make myself look foolish for spending my time trying to explain 95th-percentile to someone who is absolutely convinced that they are correct. If you like, we can post to rrd-users and you can see it explained in the "presence" of Mr Oetiker himself; though I am only subscribed to rrd-developers at present.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  24. #24
    Join Date
    Mar 2004
    Location
    Corning, NY
    Posts
    12
    jsw6, first I must say that I could cut your condescension with a knife. There are people on this board who could learn something from you if you would learn to say things in a way that didn't demean everyone.

    Second, my assumption is that the total bandwidth is actually comprised of incoming and outgoing data. Unless I'm mistaken you are indicating that the "actual transfer" is merely the outgoing transfer.

    There is a actual transfer, which is comprised of bandwidth used both in and out. There is 95th percentile billing, which is less than this amount. Then there is seperating in and out traffic, which by nature will be less than actualy, and may be less than 95th percentile as well.

    Is this not correct?

  25. #25
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by hoangelos
    jsw6, first I must say that I could cut your condescension with a knife. There are people on this board who could learn something from you if you would learn to say things in a way that didn't demean everyone.
    If you read my posts in the thread carefully, you'll note that my posts demean only the argument of the poster; and not the person. If someone's statement is wrong, I have said so; but I am interested only in the facts and statements made during the course of the discussion, not the person. If this comes off as condescending, it should be taken in context of the argument, not the participants.

    Second, my assumption is that the total bandwidth is actually comprised of incoming and outgoing data. Unless I'm mistaken you are indicating that the "actual transfer" is merely the outgoing transfer.
    Most (not necessarily all) providers bill based on the higher of inbound or outbound bandwidth used, not the sum of both. What that means in a hosting environment is that you don't pay for inbound traffic, so it's not really relavent here. Though you are correct, data is certainly transferred in both directions; I think we are only interested in what is being billed in this instance.

    There is a actual transfer, which is comprised of bandwidth used both in and out. There is 95th percentile billing, which is less than this amount. Then there is seperating in and out traffic, which by nature will be less than actualy, and may be less than 95th percentile as well.
    For the purpose of this discussion, please ignore the inbound bandwidth, as hosts don't receive bills based on that unless their provider is billing based on bidirectional traffic, which is unusual. Note that this is covered in the example chart which has been linked in this thread.

    As I have tried to explain, the 95th-percentile, billed, value for typical hosting traffic is almost always larger than the mean utilization rate, which is exactly equivilant to the "actual transfer" in either direction (you could figure 95th percentile for each direction individually if you wanted to, and we do so in our billing software). How much larger? 95th-percentile is about 42% higher than mean for typical hosting traffic.

    To clarify, the figures hosts should be interested in are total GBytes transmitted, which is equal to their mean utilization for the billing period multiplied by 2,592,000 seconds in a 30-day month, divided by 8 bits per byte (as ports are almost always metered in bits), divided by the graph's scale (kilo / mega) over giga; or 1,000,000 when going from kilobits to gigabytes.

    The 95th-percentile is not directly related to the total GBytes transferred, and you can't calculate 95th-percentile simply by knowing how many GB your box has transmitted during the month; you need the 5-minute (or whatever interval the provider uses) samples.

    Let me use another automobile example. If you have ever driven a car with a real-time Miles-Per-Gallon display, you know that when you step on the gas to accelerate onto the highway, it will read something aweful, like 8 MPG. Yet when you are travelling at an efficient speed on the highway, it may read 30 MPG or more. This is the kind of sample that is taken by port-sampling tools and graphed by MRTG or RRDtool

    When you figure your gas milage, you aren't interested in peak performance or 95th-percentile, because you pay by the gallon, like some hosting customers pay by the GByte, no matter how often they have to put their foot to the floor to speed up. You simply take the number of miles driven on the tank, divide that by the number of gallons used since your last fill-up, and you get your mean, or "average", utilization. (Note that "mean" is only one kind of statistical calculation that fits the term "average," the others are "median" and "mode".)

    Now, if you had to pay for fuel on a 95th-percentile basis, and your car's fuel utilization was sampled say every 5 minutes, you would end up with a higher gas bill because you spend a lot more of your time driving on surface streets or sitting at traffic lights than you do on the highway. All it takes is 5% of the samples to be higher than your mean MPG to end up causing you to pay more to fuel your car.

    This is a complicated subject, and I'm sorry I am not able to explain it sufficiently that everyone can understand without coming off as condescending; but it's a bit frustrating when there's a guy who is indeed incorrect, who cannot be convinced of that, and who seems uninterested in my mathematical analysis, analogies, and examples participating in the discussion. I have spent a good deal of time trying to explain this so folks can benefit from my understanding, and I hope that is appreciated.
    Last edited by jsw6; 03-29-2004 at 11:45 AM.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

  26. #26
    Join Date
    Aug 2000
    Location
    Sheffield, South Yorks
    Posts
    3,480
    Most providers I've spoken too, use highest of in/out for 95th, and if they bill on average data rate or actual usage, they use sum of in+out.
    Karl Austin :: KDA Web Services Ltd.
    UK Business Hosting and Managed Servers - Hosting for Business Users :: 0800 5429 764
    Call us today and ask about our hosting solutions.

  27. #27
    Join Date
    Feb 2003
    Location
    Rialto, CA
    Posts
    2,039
    I know of a few providers who bill 95th of sum of incoming/outgoing.

    Its really only advantageous to the provider then.

  28. #28
    Join Date
    Aug 2002
    Location
    Atlanta, GA
    Posts
    1,114
    Sorry Jeff, I guess everyone but you is wrong. I'll have to call my three primary bandwidth providers and tell them they aren't doing it right.

    Last edited by mgphoto; 03-29-2004 at 12:19 PM.
    SiteSouth
    Atlanta, GA and Las Vegas, NV. Colocation

  29. #29
    Join Date
    Feb 2004
    Location
    Louisville, Kentucky
    Posts
    1,083
    Originally posted by mgphoto
    Sorry Jeff, I guess everyone but you is wrong. I'll have to call my three primary bandwidth providers and tell them they aren't doing it right.
    They are doing it right. I have never stated that Level(3) or anyone else is doing this "wrong," only that you don't understand which value is the amount of real data transferred, and which is simply a figure that you are billed on.

    The 95th-percentile value is what comes on your bill. The amount of data really transmitted by your network or server is a function of the mean utilization, not the 95th-percentile. That's all there is left to say, and if you don't get it by now, after all the examples, all the mathematical analysis, and all the analogy I have provided; and if you still refuse to post more detailed utilization information on the customer you have who bursts to 30Mb/s for 4 hours a month, a case which even you have admitted is a cinderella case; then there is nothing more I can do to help you understand.

    What I can do is encourage other thread readers to follow my above analysis, and if it's too abstract to understand easily, think about it in automotive terms; because I'm sure everyone posting here knows how to figure out their average travelling speed or their miles-per-gallon, even if they may not already understand some of the more complex nuances of transit billing.
    Jeff at Innovative Network Concepts / 212-981-0607 x8579 / AIM: jeffsw6
    Expert IP network consultation and operation at affordable rates
    95th Percentile Explained Rate-Limiting on Cisco IOS switches

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •