Reply
Highlighted
Senior Member
Posts: 3,131
Registered: ‎07-28-2009
Kudos: 899
Solutions: 40

New Flexible framing and huge improvements in TCP latency

[ Edited ]

I have an omni Rocket AC Prism AP with 31 clients on it.  A majority of the clients are XW with some XM and only 2 AC.  Anyway we are using a system that records the TCP latency of the clients traffic.  Basically it sits in our core and records the time that ACKs take to get back to the core for the TCP traffic that flows through it.  This is not the same as ICMP latency.

 

What this does is it can show you a congested AP.  The way you can usually tell if as the usage increases the TCP latency also increases greatly for all clients you have a good indication of congestion.  On this AP as the load increased in peak hours the latency also increased to numbers at or near 150ms.

 

Well I decided to move it from old Flex to new Flex yesterday and immediatly recorded a huge drop in TCP latency for all clients.  Even when the AP was pushing traffic during peak hours at the same level as the day before the latency was around 3X less.  This was good and unexpected.  

 

I have quite a few lower quality clients on the AP and moving to new Flex saw a drop in modulation rates for most clients, but even with that it sure appears that new flex is a substantial improvment on the old Flexible framing.

 

Capture.PNGYellow line indicates change from Flex (legacy) to Flex (New)

Senior Member
Posts: 2,983
Registered: ‎07-17-2010
Kudos: 667
Solutions: 167

Re: New Flexible framing and huge improvements in TCP latency

Will be trying this tonight.
Ubiquiti Employee
Posts: 8,705
Registered: ‎04-14-2017
Kudos: 1639
Solutions: 249

Re: New Flexible framing and huge improvements in TCP latency

Neat latency results. Are you working with someone in the team already on your client modulation?
Senior Member
Posts: 3,131
Registered: ‎07-28-2009
Kudos: 899
Solutions: 40

Re: New Flexible framing and huge improvements in TCP latency


@UBNT-SNK wrote:
Neat latency results. Are you working with someone in the team already on your client modulation?

Yes @UBNT-sriram is working with me on that other issue.  It is for a client off a different AP.

New Member
Posts: 35
Registered: ‎04-27-2014

Re: New Flexible framing and huge improvements in TCP latency

[ Edited ]

@sbyrdThe problem you mentioned is about some XM CPE's getting MCS below normal? Several XM SISO clients here stay with a maximum of 39Mbps with the new flex mode but with old flex stands at 65Mbps

Senior Member
Posts: 3,131
Registered: ‎07-28-2009
Kudos: 899
Solutions: 40

Re: New Flexible framing and huge improvements in TCP latency


@gustavoghpf wrote:

@sbyrdThe problem you mentioned is about some XM CPE's getting MCS below normal? Several XM SISO clients here stay with a maximum of 39Mbps with the new flex mode but with old flex stands at 65Mbps


No I am talking about this issue: https://community.ubnt.com/t5/airMAX-AC/RX-Rates-Stable-on-Fixed-Frame-and-unstable-on-Flex-NG/m-p/2...

Senior Member
Posts: 3,131
Registered: ‎07-28-2009
Kudos: 899
Solutions: 40

Re: New Flexible framing and huge improvements in TCP latency

I do want to follow up on changes I see from Flexible Legacy to Flexable New.

 

It is highly likely you will see an increase in your CPE ICMP ping times along with ping spikes to values higher than you see on old Flex.  Some of this can be attributed to the lack of automatic HW retries on the new Flex, but can also just be what is called an admission delay.

 

Basically the way it was described to me was when you do a continuous ICMP to a CPE each ping request has to get admitted to the AP for its UL or DL frame.  If for some reason a certain ping request cannot make it into the current frame it may have to wait for its next available frame which can add some large than expected delay.

 

However for VOIP or streaming services like TCP the ACKs happen continously and the next slot can be requested from the current slot which limit the admission delay as the time between packets is much smaller than doing an ICMP ping once per second.

 

So for this AP in my example the ICMP ping times increased vs flex legacy, but as you can see the actual TCP latency dropped considerably, and in my opinion that is the type of traffic that matters most.  Just keep this in mind if you are only using ICMP latency to guage the difference between the two Flex modes.

Senior Member
Posts: 3,131
Registered: ‎07-28-2009
Kudos: 899
Solutions: 40

Re: New Flexible framing and huge improvements in TCP latency

And to show this is not a fluke here is another mixed AP with 31 clients.  A majority are XW.

 

Capture.PNGYellow line is change from Flex Legacy to Flex New.

Ubiquiti Employee
Posts: 8,705
Registered: ‎04-14-2017
Kudos: 1639
Solutions: 249

Re: New Flexible framing and huge improvements in TCP latency

Good point on ICMP vs TCP, and thanks for the data.
Emerging Member
Posts: 45
Registered: ‎05-05-2016
Kudos: 2
Solutions: 3

Re: New Flexible framing and huge improvements in TCP latency

Just curious, what software do you use for collect these datas (tcp latency)? I am searching something similar too, thanks for your share.

Senior Member
Posts: 3,131
Registered: ‎07-28-2009
Kudos: 899
Solutions: 40

Re: New Flexible framing and huge improvements in TCP latency

It's part of the Preseem platform.  You can see more on their website.

Veteran Member
Posts: 5,454
Registered: ‎07-03-2008
Kudos: 1713
Solutions: 134

Re: New Flexible framing and huge improvements in TCP latency

It's not just software -- it's a shaper/monitoring box plus SAAS https://www.preseem.com/

Reply