Mailing List Archive

hold-queue on EX4300 ?
Dear experts
I see on a customer EX4300 in VC connected to Cisco 3560 huge drop errors.

On Cisco side I've mitigated with hold-queue 4096 but on EX4300 is there
any related command ?
Or do I'm forced to change buffer size ?

Thanks in advance for your help

Cheers
James
_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Re: hold-queue on EX4300 ? [ In reply to ]
Hey James,

> I see on a customer EX4300 in VC connected to Cisco 3560 huge drop errors.

This isn't specific enough to understand what is going on.

> On Cisco side I've mitigated with hold-queue 4096 but on EX4300 is there
> any related command ?

Hold-queue is for punted packets, not particularly important or
interesting usually. So I suspect your issue with JNPR has nothing to
do with your issue in CSCO, two separate matters.

> Or do I'm forced to change buffer size ?

Maybe. Cost optimised platforms have very little delay buffer, because
it's transistor expensive, which is space expensive, which is yield
expensive. If your traffic flows are such that many ports are sending
to one port, or traffic from high speed interface is going to slow
speed interface you definitely need to tune QoS settings but it might
be that you can only improve not fix the situation.
Mind you, you'd need to do this same on 3560 too, and it's not
hold-queue but you'd assign all buffer space into a single queue, to
allow you to sink largest possible microburst.

--
++ytti
_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Re: hold-queue on EX4300 ? [ In reply to ]
Hi Ytti
we're thinking to upgrade to 10 Gbs the interface, this should help.

Cheers James

Il giorno dom 13 dic 2020 alle ore 12:45 Saku Ytti <saku@ytti.fi> ha
scritto:

> Hey James,
>
> > I see on a customer EX4300 in VC connected to Cisco 3560 huge drop
> errors.
>
> This isn't specific enough to understand what is going on.
>
> > On Cisco side I've mitigated with hold-queue 4096 but on EX4300 is there
> > any related command ?
>
> Hold-queue is for punted packets, not particularly important or
> interesting usually. So I suspect your issue with JNPR has nothing to
> do with your issue in CSCO, two separate matters.
>
> > Or do I'm forced to change buffer size ?
>
> Maybe. Cost optimised platforms have very little delay buffer, because
> it's transistor expensive, which is space expensive, which is yield
> expensive. If your traffic flows are such that many ports are sending
> to one port, or traffic from high speed interface is going to slow
> speed interface you definitely need to tune QoS settings but it might
> be that you can only improve not fix the situation.
> Mind you, you'd need to do this same on 3560 too, and it's not
> hold-queue but you'd assign all buffer space into a single queue, to
> allow you to sink largest possible microburst.
>
> --
> ++ytti
>
_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Re: hold-queue on EX4300 ? [ In reply to ]
On Sun, 13 Dec 2020 at 20:31, james list <jameslist72@gmail.com> wrote:

Hey,

> we're thinking to upgrade to 10 Gbs the interface, this should help.

It might, or it might make it worse. The absolute speed doesn't
matter, the relative speed of combined ingresses to egress matter.

If you have 10G ingress and 1G egress, you'll need lot of buffering.
If you have 5 1G ingresses pushing to 1G egress, you'll need lot of buffering.
If you have 1G pushing to 1G egress, you don't need buffering.
If you have 1G pushing to 10G egress, you don't need buffering.

I don't know which BRCM EX4300 runs, so I don't know how much buffer
it has. EX3400 is same era design and I think it's BCM565xx/firescout.

--
++ytti
_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp