[c-nsp] FW: L2 network - MPLS question

[Rick Martin]

I need to send out apologies for lack of response to my original post. After sending by question out last Friday I was called out of the office. I will be  reviewing the responses later today after playing catch up. Thanks all for the comments on this thread.

-----Original Message-----
From: Saku Ytti [mailto:saku at ytti.fi] 
Sent: Friday, May 06, 2016 12:59 PM
To: Rick Martin
Cc: cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] FW: L2 network - MPLS question

On 6 May 2016 at 19:17, Rick Martin <rick.martin at arkansas.gov> wrote:

Hey,

> We are deploying Catalyst 3650;s with ASA's at the customer sites, using Nexus 9504's for our aggregation sites. All hardware was spec'd by Cisco, this was a very big deal and they brought in MANY Cisco internal resources to come up with the design and BOM.

3650 has 12MB shared buffer, TCP window grows exponentially and ~all implementations burst (rather than pace) growth in linerate. If only single port is being used, 12MB is good enough for about 200ms RTT for 1GB. 1Gbps * 200ms = 25MB, but worst-case you'd grow the window 12.5MB, and if sender is faster, like 10Gbps, then you'd essentially see this whole burst in your buffers.
So from HW POV, 3650 isn't going to be significant limitation to your single TCP session speed, unless you want to get 1GB from China or something like that.

>  Customer experience is still poor, we have learned that using iperf or other download based speed testing tools we see a maximum of about 150Mbps download or upload speed.  If we configure the tools to use multiple threads we can see the WAN connection support full 1Gbps throughput.

64kB / 150Mbps = 3.4ms. Is the latency about 3.4ms? If so, then you'll need window scaling turned on.
--
  ++ytti