burst bandwidth

[muhsi_2015]

Hi,

I have non cisco device , the bandwidth overtime shows 4Mbps and the burst bandwidth shows 100Mbps.
there might be an issue with the device or its normal ?

Thanks

[Mark Malone]

that seems like a vey high burst rate , you could police it instead or alter the qos so it doesn't burst so high , 100mb is obviously 20 times the burst rate of whats set to reserve

[muhsi_2015]

Hi,

Is there any reason for these burst .

The device is connected to access switch

access switch-- distribution---core --fw-interent router --isp

So where do I apply qos policing

Thanks

[Mark Malone]

could be miss configured qos  , could be an application causing it , usually you apply it where the layer 3 qos is at the wan level , where you see the burst or are you seeing it somewhere else at layer 2 qos ?

you can track burst through wireshark

https://notalwaysthenetwork.com/2014/01/06/microburst-detection-with-wireshark/

[muhsi_2015]

Hi,

The output i genrated from riverbed steelcentral . 

" usually you apply it where the layer 3 qos is at the wan level ".

Traffic  captured  during the communication between two devices  in lan . 

where you see the burst or are you seeing it somewhere else at layer 2 qos ?

Actually  I have not set any qos so far , the access switches are 3850 

Thanks

[Mark Malone]

ok that's different then , it may not be qos could be just an application using more data , when you said burst I assumed you had qos set with a cir set to for burst traffic ona  router or layer 3 switch not a riverbed  , so even if its set to 4mb it can burst higher  when there is free bandwidth

so is there an actual issue here or you just see this on a chart on the management console in the rb central  , the device or whatever it maybe could be spiking traffic through the switchport itself , could be any application its using , you would need to check that locally if it is , there called microbursts from apps usually

a 3850 is unlikely maxing out at port level unless your hammering it , whats the actual local port show where this device is connected

[muhsi_2015]

Hi,

I have set  qos on wan router .This microburst can cause an issue when it flows to the wan link ? .

burst and microbursts are different ? 

Thanks

[Mark Malone]

I don't know about burst but its microburst on switch network, it can happen when qos is configured wrong or the buffers on the port where a device is directly connected pushes way more data than a port can handle in a few miliseconds this will cause a micro bursts of traffic and can spike data

you've set qos for what though the Ip address the certain type of traffic that's causing the issue ?

qos should really be planned and rolled out across all devices end to end it should not just be put on 1 interface when you see an issue

im still unsure what the problem is here is it causing issues on your wan ? the ports on a 3850 should have no problem with traffic hitting 100mb there gb ports

[muhsi_2015]

Hi,

This is a vc device , When  It is on call to  remote device  through internet , there was a lot of delay . My internet speed is only 25Mbps .

So I was assuming the microbursts was causing an issue .

" you've set qos for what though the Ip address the certain type of traffic that's causing the issue ?" 

I believe you are asking that qos is set for the problematic device ?

the answer is yes .

The qos is not set on cisco router , it is set on non cisco device 

Thanks

[Mark Malone]

Hi

 the 3850 will trust dscp markings by default and prioritise so it should have had some priority at switch level automatically on ios-xe not like the ios switches  ,but yes that can be an issue obviously coming off a gig port and bottlenecking into a 25mb wan circuit , qos should really be planned end to end between source and destination with far end too  but that may fix it for you if the far end is not under pressure that your connecting too  , what could be happening is the video works fine but then someone starts a file transfer and it hogs the wan line preventing your VC traffic getting out or slowly getting out as there is no room on the pipe causing latency

prioritizing so it gets a certain amount of traffic if its critical is a good choice , you could also enable auto qos trust dcsp or video on the VC port on the 3850 switch this will create a class and policy map which will prioritise AF41 traffic which is video traffic ,but doing it at the wan is the important part to make sure it can get out , you may not need it though but if its still an issue even after the wan qos try that

here is the link of exactly what the switch will configure when you set auto qos on a 3850

http://www.cisco.com/c/en/us/td/docs/switches/lan/catalyst3850/software/release/3se/qos/configuration_guide/b_qos_3se_3850_cg/b_qos_3se_3850_cg_chapter_0100.html#reference_1A5B72D4EB70409C8311950DF9186BDF

[Joseph W. Doherty]

If your WAN circuit provides 25 Mbps, then you should shape (not police) on your Internet router for your CIR.

Once you shape, you have the option to manage any congestion.  For starters, you might just fair-queue all your traffic.  However, if your Riverbed is wrapping all its flows into what appears to be one flow, FQ will offer less benefit.

[kp-tkr2014]

Hi,

if your Riverbed is wrapping all its flows into what appears to be one flow, FQ will offer less benefit.

Could you explain 

Thanks

[Joseph W. Doherty]

The power of FQ is each flow gets its own queue (NB: actually not, flows are hashed into a fixed number of flow queue).  There it prioritized and tail dropped independently of the other flow queues.

If Steelhead lumps multiple flows into what appears to a single flow (often done between pairs of WAN acceleration appliances) you don't obtain the benefit of different flows being placed into their (hopefully) own queue.