[j-nsp] Rate limiting

[Pavel Lunin]

Hi Timur,

2008/12/26 Timur Ibragimov <itim at ycc.ru>

> Tom Storey wrote:
>
> > A burst size of 1.5kbps as you have configured in your example only
> allows
> > traffic to increase at 1.5 kilobits each second, not a hell of a lot. At
> > that rate it would take upto 1000 seconds, i.e. 16 minutes to reach the
> > full 1.5 megabits you are wanting to supply...
> >
> I thought that burst-size-limit is measured by bytes (not 1.5kbps, but
> 1.5KB) and only important when the traffic flow is above the limit of 1500
> kbps
> according to the token-bucket algorithm.
>
> Well in my case packets begin dropping at even lower rates.


You're alright, it is measured in bytes, however the limit you set is really
low. Here is a quotation from "JUNOS Enterprise Routing" by Doug Marschke
and Harry Reynolds:

*"The setting of the burst size has always seemed to be a mystery for many
operators. Set this value too low, and potentially all packets will be
policed. Set the value too high, and no packets will be policed. The rule of
thumb is that the burst size should never be lower than 10 times the maximum
transmission unit (MTU). The recommended value is to set the amount of
traffic that can be sent over the interface in five milliseconds. So, if
your interface is a Fast Ethernet interface, the minimum is 15,000 bytes (10
* 1,500), and the recommended value would be 62,500 bytes (12,500 bytes/ms *
5).*"

As it is said in the book, burst size is *"the number of bytes that can
exceed the bandwidth constraints"*. However it doesn't mean it's how many
bytes you can transfer in addition to the the bandwidth (at least because
there is no definition of the period of time). Burst-size is not a speed of
transmission, but a size of data burst that can be transmitted with no
dropping.

--
Kind regards,

Pavel Lunin,
Senetsy,
Moscow