[c-nsp] ME3600x buffer calculations- revisited

Waris Sagheer (waris) waris at cisco.com
Fri Dec 27 03:24:32 EST 2013


Hi Julie,
Replies inline [Waris]


From: Julie Rudolph <Julie.Rudolph at cdw.com<mailto:Julie.Rudolph at cdw.com>>
Date: Thursday, December 26, 2013 2:12 PM
To: Waris Sagheer <waris at cisco.com<mailto:waris at cisco.com>>, "cisco-nsp at puck.nether.net<mailto:cisco-nsp at puck.nether.net>" <cisco-nsp at puck.nether.net<mailto:cisco-nsp at puck.nether.net>>
Subject: RE: [c-nsp] ME3600x buffer calculations- revisited

Great, thanks so much.


1)      Was the dynamic allocation of buffers to the queues/classes that needed them always there, even in 12.2x code releases?

[Waris]Yes it is always there since it is done in hardware. However queue limit was fixed in 12.2.

2)      Could you oversubscribe the 18Mbytes in code prior to any 15.x release?  I do not recall reading anything in the release notes about a change in the behavior such as this so I assume it was always there.  (Sorry if I missed something in the docs.) So, perhaps the only limiting factor to how much burst or maximum throughput a port could handle had to do with the 491520 Bytes (pre-15.24S) per queue max.  (If you couldn’t oversubscribe the 18MB, then that would come out to a max of ~38 queues that could be maxed out to 491520 bytes).

[Waris] Need to check and will get back to you.

3)      I see now how the availability of the queue percent feature allows the ability of a queue to use more than the old (old = 15.2(4)S) max of 2Mbytes provided the resources are available.   Not a question, just a comment.  And a sigh of relief.

-Julie

Julie Rudolph

From: Waris Sagheer (waris) [mailto:waris at cisco.com]
Sent: Thursday, December 26, 2013 3:38 PM
To: Julie Rudolph; cisco-nsp at puck.nether.net<mailto:cisco-nsp at puck.nether.net>
Subject: Re: [c-nsp] ME3600x buffer calculations- revisited

Hi Julie,

Replies inline [Waris]


From: Julie Rudolph <Julie.Rudolph at cdw.com<mailto:Julie.Rudolph at cdw.com>>
Date: Tuesday, December 24, 2013 4:17 PM
To: "cisco-nsp at puck.nether.net<mailto:cisco-nsp at puck.nether.net>" <cisco-nsp at puck.nether.net<mailto:cisco-nsp at puck.nether.net>>
Subject: [c-nsp] ME3600x buffer calculations- revisited

Hello:

My questions key off of this thread:  http://www.gossamer-threads.com/lists/cisco/nsp/172513.  I just seek a bit more clarification.  Wondering if anyone knows the answers to the following 2 questions:


1)      When using queue-limit percent, does this mean the switch will take that as a percentage of the 18MB of packet buffer for the ASIC serving the Gig ports on the switch?  Moreover, does it mean that the switch will dynamically allocate buffers to where they are needed, with the switch allocating the associated buffers to the busy interfaces (but not above the percentage of buffer that was specificied for that class(es) in question?

[Waris] That's correct. There is no per port buffers. Buffer allocation is done dynamically by an intelligent allocator based on the interface requirement and configuration.

2)      If I have 20 interfaces on the same asic (the one with 18MB available buffer) with QOS classes that each have the max of 2Mbytes specified as the queue-limit and those limits exceed 18MB, what will happen?

[Waris] Configuration oversubscription is supported since we assume not all the queues are oversubscribed at the same time.

3)      Lastly, has anyone ever had to configure queue-limits on the TenGig interfaces (i.e the default queue-limits weren't enough)?

[Waris] I have seen many providers who have changed the default queue-limit on 10Gig.

Thanks,
Julie
_______________________________________________
cisco-nsp mailing list  cisco-nsp at puck.nether.net<mailto:cisco-nsp at puck.nether.net>
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/



More information about the cisco-nsp mailing list