Research On High Efficient Ethernet Forwarding Algorithm For Data Center

Data Center(DC)is a cloud computing infrastructure,a huge data center can carry hundreds of thousands of servers to provide services through the network.According to the different ways of forwarding,the data packet forwarding in the data center network(DCN)can be divided into IP multicast forwarding and MAC unicast forwarding:for IP multi-cast forwarding,in the forwarding process switches need to record a large number of the multicast group information,and low-end switches storage space is limited,if the multicast forwarding information is not stored in the switch but encoded and stored in the data packet header,will bring additional traffic overhead,reduce the bandwidth utilization rate.IP mul-ticast forwarding faces scalability challenges;For MAC unicast forwarding,since the MAC address is not as easy to converge as IP address,the application of virtualization technology makes the MAC address forwarding entries growing explosively,and migration of virtual machines causes MAC forwarding tables to be updated frequently,the limited storage space of the low end switch causes the MAC unicast forwarding to face the space compression problem of forwarding table in DCN.Aiming at the scalability challenges faced by IP multicast forwarding,this paper pro-poses two efficient scalable IP multicast forwarding algorithms:IP multicast forwarding algorithm based on in-packet Lightweight Bloom Filter(in-packet LBF)and IP multicas-t forwarding algorithm based on in-packet bitmap(in-packet BMP).In the IP multicast forwarding algorithm based on in-packet LBF,the nodes of multicast forwarding tree are encoded and stored in the in-packet LBF,each intermediate switch decides to forward pack-ets by checking the Bloom Filters in the header,and the edge switch forwards packets to the local servers by checking a hash table stored in the switch.The number of nodes insert-ed in LBF is less then BF,and the last forwarding using a simple hash table is faster than using Bloom Filter.The results show that:compared with traditional in-packet Bloom Fil-ter(in-packet BF),LBF reduces the traffic leakage by 39.5%,reduces the coding length of forwarding information by 38.8%and increases throughput rate by 98.7%.For IP multicast forwarding algorithm based on in-packet BMP,the ports of switch are encoded into the in-packet bitmap.Each switch decides to forward packets from a port by checking the bitmap in the packet header.Compared to BF,the accurate forwarding of BMP avoids the addition-al traffic overhead brought by false positive forwarding,more over the efficient cuckoo hash lookup improves forwarding throughput rate significantly.The results show that:compared with the BF,the traffic leakage rate and code length can be reduced by an average of 4-6 times using BMP,and the forwarding throughput rate is also better than BF.For the forwarding space compression problem that MAC unicast forwarding faces,this paper proposes a reusable fixed length coding(RFLC)which supports fast update.The coding method chooses different fixed length parameter for different scales of MAC for-warding tables,and the fixed length parameter corresponds to the fixed length code set,and different MAC address items reuse the code in the set,thus the coding method can select the appropriate fixed length parameter to shorten the length of the coding,and this method use the cuckoo hash to increase forwarding rate.The results show that:compared with the traditional fixed-length encoding and conbined huffman encoding,the code length reduced by 1.03%-27.53%,and the throughput rate increased by 34.73%-59.59%.Moreover the traditional coding methods are static,can not support the dynamic update of the forwarding table,but RFLC is flexible to support dynamic update of MAC unicast forwarding table.