Description of the problem
I. OSN3500 device ping test results on the current network:
1, EFT8----EFT8 single board 2M channel can ping through 64000 packets
2、EFT8A----EFT8 single board 10M channel can ping through 22000 packets
3、EFT8A----EFT8 single board 20M channel can ping through the 25100 packet
4、EFT8A----EFT8 single board 42M channel can ping through 36000 packets
Second, because EFT8----EFT8 single board 2M channel can ping through 64000 packets, while EFT8A----EFT8 single board 42M channel can ping through 36000 packets, suspect EFT8A single board performance problems.
Processing
I. Increase bandwidth ping test
Result:
1, EFT8A ----EFT8 single board 10M channel can ping through 22000 packets
2、EFT8A----EFT8 single board 20M channel can ping through the 25100 packet
3、EFT8A----EFT8 single board 42M channel can ping through the 36000 packets
Replace the single board with a larger cache for ping test.
Result: EFT8----EFT8 single board 2M channel can ping through 64000 packets.
Root Cause
First of all, analyze the process of PC ping large packets, PC machine can ping the longest 65500 bytes, and the default MTU value of the PC for 1500 bytes, so ping more than 1500 messages need to be sliced, then 22000 bytes of the packet about how many slices to be divided into how much 22000/1500 = 14.6, that is to say, if you use the PC ping 22000 byte packet, PC will split the packet into 15 pieces and send it out.
Second, using the IP address of 10.169.54.135 PC ping another IP for 10.169.46.112 PC, in the process of ping packet capture, found that the actual situation and the above calculations are consistent with the results, the computer sent a total of 15 slices, of which 14 slices of the length of 1514 (each slice of the packet header needs to include the content), 1 The computer sends a total of 15 slices, 14 of which are 1514 (each slice needs to contain the contents of the packet header), and 1 slice is 1322 (one slice is less than 1500 after being cut into 15 slices).
Third, check the packet capture results in the message sending time, the first slice issued by the time is 2015-04-14 20:03:48.251111, the last slice issued by the time is 2015-04-14 20:03:48.255325, the time difference is 0.004214 seconds, the amount of data in 15 slices: 1514 * 14 + 1322 * 1 = 22518 (Here the actual data sent by PC is greater than 22000 of ping because each slice needs to add the packet header, so the actual data is going to be larger than ping, and the longer the packet length the more slices there are, the more extra data there is.)
Four, then this time pc ping packet sending rate can be calculated: 22518 * 8 / 0.004214 = 42.748Mbit / s. You can see ping large packets caused by the burst is very serious, although only ping a 22000 bytes of the packet, but the PC in the time of 4ms to send all of these slices, resulting in The instantaneous burst of traffic reached about 42.7M, and at this time our bandwidth is only 10M, the bandwidth is far less than the burst of traffic, so this time you need to temporarily send out the data is not sent to the single-board cache to be sent later, if the single-board cache is not enough may lead to loss of packets, the performance of the phenomenon is the ping can not be.
Fifth, in this case, how much cache is enough? From the above data, we know that we need to forward 22518*8=180,144bit data in 0.004214 seconds, but our bandwidth is only 10Mbit/s, so in 0.004214 seconds can be forwarded out of the data in 0.004214*10,000,000bit=42140bit, which still leaves 180144bit-42140bit=138,004bit=17.250KB. And our N2EFT8A single-board port cache is only 20KB (N1EFT8 cache is 64KB), so just enough to pass the 22000 message (the actual test at this time is just enough to pass, but then larger on the not work), if the length of the message is further increased, it may lead to insufficient cache loss. If the length of the message is further increased, it may lead to insufficient cache to lose packets.
The solution
Configure the shaping traffic shaping function on the devices (switches) that are connected to the transmission to ensure that the traffic entering the transmission is smooth and without bursts.
Replace the single board with a larger cache and increase the bandwidth (not recommended).
Recommendations and Summary
First, the data veneer are rigid pipes without any elasticity, any moment as long as the traffic exceeds the bandwidth can only be discarded. And transmission is like a highway is mainly used for forwarding large amounts of data, not storage, so the highway will only be designed with a limited number of emergency parking lanes (caches), so the problems encountered in the burst need to eliminate the burst from the source, rather than relying on constantly increasing the transmission cache to solve the problem. So the test method of using PC ping large packets is unreliable, to verify whether a rigid pipe can provide the nominal bandwidth need to use a professional meter to verify (such as the meter: SmartBits).
Second, in the actual network may be due to the reasons of the data source (such as sending large packets), after the transmission of large packets by the data source slice, resulting in large fluctuations in the flow of the formation of bursts resulting in instantaneous traffic over the transmission bandwidth caused by the loss of business packets, at this time, as long as there is a slice of the loss of the entire packet at the receiving end can not be reorganized, the performance of ping packets do not work. If there is such a large packet burst scenario in the existing network, it is recommended to configure shaping traffic shaping function on the device (switch) that interfaces with the transmission to ensure that the traffic entering the transmission is smooth and without bursts, or replace the single board with a larger cache (not recommended).
The technical information related to this chapter and the troubleshooting procedures for SDH equipment are provided by Shenzhen Optical Transmission Network Technology Co. Huawei SDH Optical Transmission Equipment,SDH Transmission Equipment Sales


Chinese
English





