Hi Guys,
I would like to share with you an interesting case that i encounter on S2700 SI series. I will skip any introduction, and i will show first the topology.
Few months ago, our customer was using 100M ethernet uplink line to connect access switch 2700 with distribution switch S5700. In order to increase the uplink bandwidth, the solution was changed from 100Mbps line to 1 Gbps line as the above picture shows.
Unfortunately this adjustment was not a good one, because immediately after change, the interfaces that were connecting access users, started to increment discards for outbound direction. No discards were spotted on uplink line, nor inbound nor outbound.
Ethernet0/0/1 current state : UP
Line protocol current state : UP
Description:host1
Switch Port, PVID : 1105, TPID : 8100(Hex), The Maximum Frame Length is 1536
IP Sending Frames' Format is PKTFMT_ETHNT_2, Hardware address is d4b1-10cd-870f
Port Mode: COMMON COPPER
Speed : 100, Loopback: NONE
Duplex: FULL, Negotiation: ENABLE
Mdi : AUTO
Last 60 seconds input rate 103712 bits/sec, 80 packets/sec
Last 60 seconds output rate 1760720 bits/sec, 175 packets/sec
Input peak rate 163903392 bits/sec, Record time: 2014-05-19 20:26
Output peak rate 274790504 bits/sec, Record time: 2014-05-19 20:26
Input: 7852997 packets, 1499049380 bytes
Unicast : 7852997, Multicast : 0
Broadcast : 0, Jumbo : 0
CRC : 0, Giants : 0
Jabbers : 0, Fragments : 0
Runts : 0, DropEvents : 0
Alignments : 0, Symbols : 0
Ignoreds : 0, Frames : 0
Discard : 0, Total Error : 0
Output: 12833895 packets, 16448124820 bytes
Unicast : 12815530, Multicast : 18363
Broadcast : 2, Jumbo : 0
Collisions : 0, Deferreds : 0
Late Collisions: 0, ExcessiveCollisions: 0
Buffers Purged : 0
Discard : 12035, Total Error : 0
Input bandwidth utilization threshold : 100.00%
Output bandwidth utilization threshold: 100.00%
Input bandwidth utilization : 0.10%
Output bandwidth utilization : 1.76%
Actually, even if you don't believe me, this graph made me understand what really happens.
1. discards appear immediately after change 10 June (uplink swap)
2. the event incidence is not periodical.
3. the quantity is never the same.
This 3 points made me think we are facing a problem with burst traffic handling. Obviously a 1Gbps interface will be able to receive a higher amount of burst traffic, while mixed incidence and quantity are conditions for burst traffic type.
Because now we know the root-cause, how to solve this?
1. Enable the burst traffic mode to improve the capability of the device for carrying the burst traffic.
<Quidway> system-view
[Quidway] burst-mode enable
By default, to ensure that each interface uses buffer resources of the system fairly, the buffer resources of the system are evenly allocated to the interfaces. In this case, the packet loss threshold is low. If the network traffic is even and stable, you can use the configuration. If the burst traffic is heavy on the network, packets are lost easily. After the burst traffic mode is enabled, if the burst traffic occurs on a single interface, other interfaces can use idle buffer resources of the system. The capability of the device for carrying burst traffic is improved.
Well this is a cool command because the situation was improved, let's check the graph now:
We can see that after 18th of June, discards are insignificant.
2. We almost solve it. But how to make discards disappear once and for all?
Below picture speaks for itself.
I hope you enjoy reading this case.
Bye