95th percentile 是什么意思？

gdtv

95th percentile refers to a method of bandwidth tallying, most commonly associated with colocation. In 95th percentile billing, the customer is not charged for the top 5% of bandwidth used in any given month.

谁来帮忙翻译一下

pigicoffee

也就是说 他的显示2M 但是你实际可以用的是2M/95%

toapple

英文不好C大呢？

cpuer

比如你一个时间段的带宽占用是这样的，以某台Krypt的服务器为例，

Snap2.jpg (72.69 KB, 下载次数: 12)

2009-11-10 10:21 上传

点击文件名下载附件

上图是1分钟一次记录的，所以把所有1分钟一次的记录排在一起后，比如一个小时会产生60个带宽占用值，进入服务器带宽值60个，服务器出去的带宽值60个。

然后分别对这2个60个做排列，从高到低排列，去掉5%（就是前3个高值），剩下57个里面的最高值（第四个高值）就是95%th的带宽值，希望你能明白。

这样就可以明白图右下角的95%的inbound和outbound的意思了，得到的是一个小时的，红线就是outbound的95%th值坐标，黄线就是inbound的95%th坐标。

如果要一个月的就是把一个月的拿来进行排列就行了。

pigicoffee

The 95th percentile is the smallest number that is greater than 95% of the numbers in a given set. The reason this statistic is so useful in measuring data throughput is that it gives a very accurate picture of the use and cost of the bandwidth. We take a snapshot of your bandwidth usage every 5 minutes throughout the billing period (usually 30 days). These 8,640 snapshots are then used to create the 'set' mentioned above. We then employ the 95th percentile rule to calculate your bandwidth usage and then bill accordingly if necessary.

The advantage to you as the customer is that you get the performance of a high-speed connection, while paying only for actual usage.

In the example above, we would take out the higher 5% of the Max In/Max Out bandwidth measurement from the monthly MRTG graph which is 884 Kbps.

We would then take the bottom 94% of the remaining samples taken, in this example that is 337 Kbps.

Using this method each 1,000 Kbps is 1 Mbs per month. The percentage of bandwidth per Mbs would be x/1000 or 337/1000 = .337 Mbs or 107 Gb per month.

lemss

那也太垃圾了吧

pigicoffee

我没有说明白看我后来的解释

cpuer

我在四楼给中文解释了，以服务器实图为例了。

gdtv

看了还是有点不明白，是不是说：95%的时间里带宽不能超过2mbps ?

在VPS上装什么软件可以看到4楼的图？

cpuer

因为时间秒不好做图，所以都是多少分钟平均取一个带宽占用值。
没有一个准确描述的句子，95%的时间（几分钟取一次平均）里带宽不能超过2mbps，大体上上是这样的。

这个图片是krypt的交换机监控的流量图。

如果是VPS的话只能自己按照软件比如mrtg来监控了，大致类似。