Mailing List Archive: rate limiting question

rate limiting question

Dec 8, 2003, 6:20 PM

Post #1 of 4 (896 views)

Hi folks,

Got sort of a basic question in regards to rate limiting. I'm trying
to figure out a good method of determining what to set the burst to.
The docs say that a decent measure is 10x the MTU but I'm not certain
thats the best method. (ie, limiting someone to 20 megs vs 50 megs, the
burst period etc would necessarily vary, right?) So what sort of best
practice do people out there use for their setups?

-Wayne

---
Wayne Bouchard
web@typo.org
Network Dude
http://www.typo.org/~web/

rate limiting question [ In reply to ]

harry at juniper

Dec 8, 2003, 6:40 PM

Post #2 of 4 (882 views)

Permalink

Another rule of thumb, especially for high-speed links (> OC-12c) is to set
the max-burst to a value that represents 3-5 milliseconds of line rate
transmission.

As you point out, a 20 M vs a 50 M limiter may well use the same burst size
parameter. The main concern is setting too small a burst, which can make the
policer seem broken because throughput is measured below the configured
bandwidth limit.

HTHs

> -----Original Message-----
> From: juniper-nsp-bounces@puck.nether.net
> [mailto:juniper-nsp-bounces@puck.nether.net] On Behalf Of
> Wayne E. Bouchard
> Sent: Monday, December 08, 2003 3:20 PM
> To: juniper-nsp@puck.nether.net
> Subject: [j-nsp] rate limiting question
>
>
> Hi folks,
>
> Got sort of a basic question in regards to rate limiting. I'm
> trying to figure out a good method of determining what to set
> the burst to. The docs say that a decent measure is 10x the
> MTU but I'm not certain thats the best method. (ie, limiting
> someone to 20 megs vs 50 megs, the burst period etc would
> necessarily vary, right?) So what sort of best practice do
> people out there use for their setups?
>
> -Wayne
>
>
> ---
> Wayne Bouchard
> web@typo.org
> Network Dude
> http://www.typo.org/~web/
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/junipe> r-nsp
>

rate limiting question [ In reply to ]

ras at e-gerbil

Dec 9, 2003, 12:48 AM

Post #3 of 4 (876 views)

Permalink

On Mon, Dec 08, 2003 at 03:39:56PM -0800, harry wrote:
> Another rule of thumb, especially for high-speed links (> OC-12c) is to set
> the max-burst to a value that represents 3-5 milliseconds of line rate
> transmission.
>
> As you point out, a 20 M vs a 50 M limiter may well use the same burst size
> parameter. The main concern is setting too small a burst, which can make the
> policer seem broken because throughput is measured below the configured
> bandwidth limit.

Is there some particular reason why there isn't an automatic setting
available for the max burst? :)

--
Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)

rate limiting question [ In reply to ]

harry at juniper

Dec 9, 2003, 10:39 AM

Post #4 of 4 (879 views)

Permalink

It seems that a default, which could be configured explicitly, would be
cool. No idea if there are any plans for such an enhancement.

> -----Original Message-----
> From: Richard A Steenbergen [mailto:ras@e-gerbil.net]
> Sent: Monday, December 08, 2003 9:48 PM
> To: harry
> Cc: 'Wayne E. Bouchard'; juniper-nsp@puck.nether.net
> Subject: Re: [j-nsp] rate limiting question
>
>
> On Mon, Dec 08, 2003 at 03:39:56PM -0800, harry wrote:
> > Another rule of thumb, especially for high-speed links (>
> OC-12c) is
> > to set the max-burst to a value that represents 3-5 milliseconds of
> > line rate transmission.
> >
> > As you point out, a 20 M vs a 50 M limiter may well use the
> same burst
> > size parameter. The main concern is setting too small a
> burst, which
> > can make the policer seem broken because throughput is
> measured below
> > the configured bandwidth limit.
>
> Is there some particular reason why there isn't an automatic setting
> available for the max burst? :)
>
> --
> Richard A Steenbergen <ras@e-gerbil.net>
> http://www.e-gerbil.net/ras
> GPG Key ID: 0xF8B12CBC (7535 7F59
> 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)
>