You could end up in a DoS situation where some rouge source send 1gbps worth of packets matching your priority class. That's just of top of my head. Also, depending on what you decide to match on (e.g IP, DSCP marks, ingress iface, etc) you'll constantly will have to tinker with the policy whenever you need to add/remove traffic which will be painful (unless you automate it, of course) depending on your change control policies and frequency of changes. If this is a long haul link, one can try to build a business case between cost of additional circuit (or upgrading the existing one) and efforts to design and operate. If this is in the data center, just add another port, use ECMP and forget about QoS. That's just my opinion. As they say: "YMMV"
----------- Sent from mobile > On Nov 8, 2016, at 09:58, Aaron <aar...@gvtc.com> wrote: > > Thanks Andrey, you mentioned ... "In fact, I found in some cases QoS hurts > more than it helps, because you are limiting the amount of bandwidth > available to high priority packets (either priority, policy or shape) and > you will have a chance for tail drops or WRED even on a non congested link." > > Question - what if you don't limit the amount of bandwidth available to high > pri traffic, in other words, what if, on a 1 gig link, you say 1 gbps for > the priority queue and 100% available to everything else, and then if there > is hi-pri traffic, it gets what it needs, but when there isn't hi-pri > traffic taking up 1 gbps, then everything else can use up to 100% minus what > hi-pri traffic needs. ? > > eng-lab-3600-1#sh run policy-map qos-test > Building configuration... > > Current configuration : 126 bytes > ! > policy-map qos-test > class qos-test > priority > police 1,000,000,000 > class class-default > shape average percent 100 > ! > End > > - Aaron > > p.s. sorry I don't have a junos example, but I've been tinkering with this > in Cisco IOS > p.s.s. commas inserted into the police command for ease of readability for > y'all on the mail list > _______________________________________________ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp
