[j-nsp] Burst Reset question on Juniper M5

Guy Davies aguydavies at gmail.com
Fri May 16 05:37:48 EDT 2008


Hi,

Michael's basically right.  If you imagine that you have a bucket that
is sized for 10xMTU of the link (e.g. 15000 Bytes) and the link is a
1Mbps link, it would only take 15*8/1000 seconds (120ms) to totally
empty a full burst buffer.  If you have a transmit-rate of 1kbps on
your 1Mbps link, then you'd take 120s to empty your buffer completely
if you received no more packets during that 120s!

So, let's look at a slightly more realistic situation.  You have a 1GE
interface but you only wish to transmit at 10Mbps on a particular
queue.  So, you set your transmit rate to 10Mbps.  That is the size of
the hole in the bottom of the bucket.  You set your burst size to be
10 MTUs (we'll assume that the MTU is 1500 *Bytes*).  That is the size
of the bucket.

So, you can transmit 10000000 bits per second onto the link.  Your
buffer is 120000 bits in length.  Therefore, you can empty your buffer
(assuming it starts completely full and no more traffic arrives) in
12ms.  If your buffer is already full and you continue to receive
traffic into the buffer at 10Mbps, you will never empty the buffer but
nor will you tail drop any packets (theoretically - you may drop if
your traffic is not absolutely consistent and a packet arrives before
there is sufficient space vacated by the packet at the front of the
queue).  If for any period, the traffic rate drops below 10Mbps
arriving into the queue, the fill level of the queue will drop.  If
the queue is full, it will take at most 12ms for a packet to get from
the back of the queue to the front.

That's a very simplistic view of what happens with a leaky bucket queue.

Rgds,

Guy

2008/5/15  <michael.firth at bt.com>:
> I've seen it explained as a 'leaky bucket' model -  the bandwidth limit effectively becomes how fast data is allowed out of the bucket, and the burst-size-limit becomes the bucket size.
>
> I'm not sure if this is an accurate representation of how it is really modelled, then, if it is, you would have to wait 10 seconds between each 10Mbit burst on your 1Mbit/s limited link. If you burst at the high rate for too long, the bucket becomes full and packets start being discarded.
>
> Hope this helps
>
> Michael
>
> ________________________________
>
> From: juniper-nsp-bounces at puck.nether.net on behalf of Dave D
> Sent: Wed 14/05/2008 18:41
> To: juniper-nsp at puck.nether.net
> Subject: [j-nsp] Burst Reset question on Juniper M5
>
>
>
> Hello all,
>
>
>
>  So a customer of mine has been asking me about burst-size-limit on Juniper
> routers. Before I begin, I have a Juniper M5 running 7.6R4.3. My customer
> has the burst size limit set on his port, but he is asking me some questions
> I can't find the answers to.
>
>
>
> This was a page from Juniper's site explaining burst-size-limit's...
>
>
>
> http://www.juniper.net/techpubs/software/junos/junos76/swconfig76-network-interfaces/html/interfaces-summary53.html
>
> My customer is asking about the conditions for the RESET of the burst
> counter...
>
>
> " its simple enough that if you go from 1 to 10mbps on a 1mbps
> limited connection, and are allowed 1 Gig of traffic, it starts counting
> at 10mbps, and when it hits 1 gig, it shapes to 1Mbps. That's easy.
>
> But what if you burst to 10Mbps for 200 megs, then drop to 1Mbps for
> 10 seconds, then burst to 10Mbps for another 300 megs, drop to 1mbps,
> burst again, etc - you could end up with an aggregate equivalent bandwidth
> rate of 9Mbps for eg even tho your shaping is at 1mbps.
>
> I'd expect there's some counter or multiplier or something that
> determines how long you have to *NOT* burst for before the counter is
> cleared (or some decay rate or something, perhaps every non burst
> byte sent comes off the burst counter, that'd be one way.
>
>
>
>
>
> I would think Juniper would be smart enough to have some sort of built in
> reset but I am no certain. Does anyone know offhand?
>
>
>
>
>
> Any and all explanations would be greatly appreciated,
>
>
>
> Thank you very much everyone.
>
> Dave
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/juniper-nsp
>
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/juniper-nsp
>


More information about the juniper-nsp mailing list