Development of the Network in Datacenter
Along With the developing technology in data centers, the network infrastructure must be developed to meet the needs. Until today various technologies have been used to meet the needs. These network technologies respectively are as follows.
STP(Spanning Tree Protocol),RSTP and MST
VPC, Mc-LAG, VSS, and MCEC
FabricPath
VXLAN&EVPN
At the moment the most popular technology in Datacenters is EVPN&VXLAN. Before explaining the technical details, we will shortly explain why STP or other technologies are inadequate. Then we will examine the use of Vxlan in the data center.
STP, RSTP, and MSTP;
To form a loop-free structure with STP is hard. Due to the nature of Layer2 packets, it does not have a TTL value like IP packets. As topologies grow, network topologies that are not properly configured with STP are open to loop. Especially if the environment is dynamic like Datacenters and new devices are added and removed every day, things get out of hand. If a standard is not specified for customers who manage their systems in the Datacenter, the installations they will make with their standards will complicate things thoroughly. Network outages cause companies to lose time and of course money. Establishing a topology with STP at high traffic speeds can cause disaster.
If we explain with a simple example. Since the convergence time of Stp is high, when link traffic reaches values such as 10G, 20G, 40G, and, 100G from 1G, it will cause significant data loss that cannot be ignored. If the link speed is 40Gbit/sec at the link and convergence time is 10sec. When we write it in bytes, it is 400/8 = 50GB. It causes a data loss of approximately 50GB in 10 seconds.
10seconds is given for example. The convergence time is between 30 sec and 50 sec in STP. This period can be reduced with the improvements to be made. With the use of RSTP and MSTP, these times can be under 1 second. even subsecond convergence times can result in significant traffic drop at the high-speed link. However, as a result, using STP does not eliminate the possibility of blocked ports and loops in the environment. Fine-tuning is required to use redundant links in STP topologies. However, even if we make these fine adjustments, we cannot bring the convergence times to the desired levels. However, as a result, using STP based on these technologies does not eliminate the possibility of blocked ports and loops in the environment. Fine-tuning is required to use redundant links in STP topologies. However, even if we make these fine-tuning, we cannot bring the convergence times to the desired levels.
Various issues make the adoption of STP for a large data center network, difficult. Some of the major issues include the following:
Convergence issues
Unused links
Network Scale
Lack of dual-homing support
Traffic storm issues
Suboptimal forwarding
Lack of ECMP
VPC,MC-LAG,MCEC and VSS:
The main goal of these technologies is to use links as active-active these technologies allow a downstream device (host or switch) to attach to pair of switches. The figure shows how a typical STP network can be made more efficient with vPC. The downstream device is attached to both vPC peers using a regular Port-channel or Etherchannel configuration. Both links are in an active state thereby allowing active-active forwarding.
Multidestination traffic to the downstream device is sent via only one of the peers, thereby avoiding duplication.
While vPC, MC-LAG, MCEC, and VSS addresses some limitation of STP, it remains limited to a pair of switches.
FABRICPATH:
FabricPath is Cisco's proprietary, Cisco FabricPath is ad MAC-in-MAC encapsulation that eliminates the use of STP in Layer2 networks. It uses Layer2 Intermediate System (IS-IS)
with appropriate extensions to distribute the topology information among the switches that are part of the network. So the switch act as a router. FabricPath networks use Intermediate System-to-Intermediate System (ISIS) and CE(Classical Ethernet) networks use STP to construct a forwarding topology. The vendors use their different technologies. Juniper Qfabric,Brocade VCS etc.. Combining the systems that do not share the common standards are not possible. Not using common standards extends business processes. As a matter of fact, there is confusion among the manufacturers in this part. As time passed, while everyone waited to meet the common standard the networking vendor keep continued to develop their technologies.
Today, with the virtualization of systems, the importance of flexible infrastructures has increased. Environments have become more dynamic. Fabricpath offers a more suitable environment for expansion. Systems are now required to be plug and play, This flexibility has been achieved in datacenters established by combining Fabricpath, vPC +, and FEX features of Nexus. Cisco, as well as a lack of multivendor support. Also, IP is the de facto standard in the networking Industry. IP-based encapsulation has become a necessity. With the development of routing capabilities of switches, which we call edge or TOR (Top of rack), and their becoming stronger in terms of hardware, there is no obstacle to installing an IP-based fabric. At this point, Evpn & Vxlan comes up. The following topology is set up with farbricpath. As can be seen, there is no restriction in the Spine part as in Vpc or MC-LAG. It is open to horizontal growth.
Also, if you wonder Benefits of Cisco FabricPath in the Data Center. there is a nice table about comparing with vPC, STP in the following table which taken from Cisco web page.
コメント