TCP Output drop at MPLS Circuit 40Mbps

[Ken Lee]

Hi

I have an international circuit and found that the TCP output drop has consistently increased bit by bit. (Screenshot as attached).
I have checked layer 1 is fine, what could be the issue besides this?

Thanks.

ken

[SOHAN HEGDE]

Hello Lee,

It seems to be case of over utilization of CCT which is causing output drops, As  Input rate and output rate for last 5 mins are reaching 36 Mbps and 34Mbps respectively, Also Input queue is increasing.

Please rate the post if it is helpful

[Ken Lee]

Hi Sohan

Thanks for your input post. As I checked from the bandwitdh, it is within the max bandwidth , any where I can confirm this be it in the command used in the router or other places..?

p/s: btw the Bandwitdh is 50Mbps.

Thanks.

[SOHAN HEGDE]

Hii Lee,

Noted that Bandwidth is 50Mbps, However, total drops which you are observing is historic data and there is possibility that traffic has reached 50 Mbps in past few days.

If you are using QOS in WAN interface then you can check with command '"sh policy-map interface 'WAN interface-name' | i Service-policy|Class-map:|drops/"

it shows if there is a policy-exceed drops (which will occurs only in case of over utilization).

sh policy-map interface 'WAN interface-name' | i Service-policy|Class-map:|drops

  Service-policy output: COS-OUT-1536K
    Class-map: class-default (match-any)
      Service-policy : COS-OUT-1536K
        Class-map: DSCP-OUT-DATA (match-any)
        (depth/total drops/no-buffer drops) 0/6773/0   <<< Drops in DATA traffic due to over utilization

Please rate the post if it is useful

[Ken Lee]

hi Sohan

Thanks, but we are not using QoS from the router side. So can I say that this is the only way to check for the traffic if it is being dropped by over-utilisation?

Thanks.

[SOHAN HEGDE]

Hello Lee,

You can ask for utilization graph from ISP provider for past 1 week, most ISP will definitely provide you .

Please rate the post if it is helpful

[Ken Lee]

Hi Sohan

Thanks again for your reply. I have the graph BW as attached, but from your expertise, do you think this is over-subscribed with the traffic in and out given in the graph?

Regards,

[SOHAN HEGDE · Accepted solution]

Hii Lee,

Looking at the graph you shared, yes i can confirm that CCT has reached peak multiple times . generally  traffic showing more than 80% is consider as over utilized. Depending upon graph polling interval it can miss burst for short interval which leads to drops.

[Ken Lee]

hi Sohan

It is set to 60secs for the polling interval. In any case , thanks for your valuable input and I have rated your reply.

ken

[Joseph W. Doherty]

Most likely your drops are egress queue overflows during transient congestion.

You bandwidth stats, on the interface, and graph stats are basically average utilization over longer time periods (e.g. 5 minutes, the default, for the former).  Queue overflows can happen very rapidly at Mbps bandwidths.

As I notice your interface is physically running at 100 Mbps, but you say your circuit is 50 Mbps, you might also be having even more drops within the provider's path.

If this is an international circuit, and bandwidth is 50 Mbps, your default queue allocation of 40 packets is likely undersized.  (Research BDP [bandwidth delay product]).

If your path's bandwidth is less then your interface bandwidth, I would suggest you shape for the former.