[j-nsp] forwarding class nd loss priority
df at webkom.se
Fri May 1 18:21:08 EDT 2009
Then use my last mail and calculate the buffer size of 20% of the value
I gave you. Then en 25% fill level and drop prob 100% is ok!
David Ball wrote:
> Fill level refers to the level of 'fullness' of your buffer.
> Depending on what hardware you're using, the max size of the buffer
> will differ. At any rate, you've allocated 20% of your hardware's max
> buffer size for your buffer size for that scheduler.
> So, if that scheduler's buffer gets 25% full, you'll start dropping
> traffic from that queue.
> On 01/05/2009, Andrew Jimmy <good1 at live.com> wrote:
>> Hi Patrik,
>> Many thanks for the reply. It really helps a lot. Can you please put some
>> comments on the following configuration. Just to avoid jitter, I'm keeping
>> the voice queue small.
>> fill-level 0 drop-probability 0;
>> fill-level 25 drop-probability 100;
>> transmit-rate percent 20 exact;
>> buffer-size percent 20;
>> priority high;
>> drop-profile-map loss-priority any protocol any drop-profile voice-profile;
>> Is this good on Gig interface for Voice. What is exactly this file-level
>> with drop-probability.
>> -----Original Message-----
>> From: Patrik Olsson [mailto:df at webkom.se]
>> Sent: Friday, May 01, 2009 10:26 AM
>> To: Andrew Jimmy
>> Cc: juniper-nsp at puck.nether.net
>> Subject: Re: [j-nsp] forwarding class nd loss priority
>> schedulers and schedulers-map is to schdule traffic egress.
>> The scheduler decides how much bandwidth a forwarding class gets.
>> Ingress forwarding class is decided and also a drop priority ion case
>> the traffic needs to be dropped.
>> In the scheduler within a forwarding class different drop-priorities can
>> be treated different through applying different drop profiles.
>> When it comes to voice, it should have it's own forwarding class of a
>> high priority. It should not be using different drop priorities and
>> different drop profiles. It should have one that says: drop everything
>> if you need to queue. Because voice traffic dont like to be queued and
>> sent later. It will mess the sound up. In return voice traffic should
>> have the highest priority and whatever bandwidth it need to push any
>> other traffic away.
>>> Forwarding class and loss priority is some of the confusing stuff in JUNOS
>>> world. Can someone write about loss priority/scheduler maps in detail
>>> with a example. Like what if you are running MPLS VPN and you want to keep
>>> priorities voice traffic (if you want to avoid jitter for the sip call)
>>> weather the congestion is happening or not. How do you do it practically.
>>> fill-level 0 drop-probability 0;
>>> fill-level 25 drop-probability 100;
>>> Is this good to have on Gig interface for Voice traffic.
>>> transmit-rate percent 20 exact;
>>> buffer-size percent 20;
>>> priority high;
>>> drop-profile-map loss-priority any protocol any drop-profile
>>> I will highly appreciate if someone can dig into this along with
>>> juniper-nsp mailing list juniper-nsp at puck.nether.net
>> +46 (0)709 35 22 99
>> +46 (0)8 559 26 488
>> juniper-nsp mailing list juniper-nsp at puck.nether.net
+46 (0)709 35 22 99
+46 (0)8 559 26 488
More information about the juniper-nsp