VCF vertical virtualization technology architecture

VM and its migration drive the development of a large-scale L2 network in the data center. With the expansion of the network scale, the number of network devices increases, and network management becomes a thorny issue in the infrastructure management of the data center. At the same time, the modern Big Data Center also puts forward higher requirements on the port density that the network provides to the server. For example, the scale of thousands of servers has become a common demand in the reality of Internet data centers. As an effective means to increase the port density of access devices, port expansion technology has gradually become more mature and recognized by the industry. VCF Vertical virtualization technology Vertical Converged Framework, Vertical integration Framework, hereinafter referred to as VCF) is an implementation method of this technology, to meet the needs of data center virtualization and high-density access and simplify management. Similar Cisco technologies are FEX.

VCF supports heterogeneous extension of the system in the vertical dimension, that is, a box-type device is added as a remote interface board to the main device system on the basis of a logical virtual device, to expand the I/O port capability and perform centralized control management. To facilitate the description, we will compare the establishment and management process of vertical VCF with the traditional horizontal functions of IRF.

IRF horizontal) stack topology mainly includes two types: chain and ring. Devices can be divided into Master and Slave by role. Under certain conditions, the Server Load balancer instance can be converted to a Master instance. The service processing capabilities of the two instances are the same level, but the Server Load balancer instance is in the "neither nor" status.

For VCF (vertical), devices are divided into CBControlling Bridge and PEPort Extender by role. CB indicates the control device, PE indicates the vertical expansion device, that is, the port extender or remote interface board ). Generally, the PE device is not capable of acting as CB, and the Management topology is difficult to be upgraded. It is in the "Neither neither nor" state.

1. The left side is the box-type device or the box-type device forming an IRF stack horizontal virtualization system, which has two topology forms: The Ring stack and the chain stack dotted line; the right side is a VCF vertical virtualization System (VCF Fabric) formed by frame devices and box devices. To facilitate comparison, CB is composed of IRF stacks.

Figure 1. Comparison between IRF horizontal virtualization and VCF vertical Virtualization

Generally, for IRF horizontal stack), the control plane is managed by the Master, and the forwarding capability and port density increase with the increase of Slave. For VCF vertical) Fabric, the control plane is managed by the Master in CB or IRF. The port density increases with the increase of PE, but the overall forwarding capability still depends on the CB device.

VCF can be used in combination with IRF technology. The system has the advantages of single management point, cross-device aggregation, and plug-and-play. It also enhances the scalability of vertical ports.

1. VCF Technical Mechanism

For VCF, the CB role can be assumed by box-type devices with strong processing capabilities, or horizontal stacking Based on IRF technology. PE is generally a low-cost box device. In practice, the CB role is mostly a horizontal stack, which is beneficial to PE uplink redundancy. The following technical instructions are based on this.

1. Topology Management

In Figure 2, the CB role is a typical IRF stack. The PE role is a cartridge device. The CB and PE interconnection ports are called vertical Fabric ports. Vertical Fabric ports are logical interfaces that can be composed of one physical port or multiple physical ports. Dedicated cables or optical fiber connections can be used between CB and PE.

Figure 2. Typical VCF Topology

The PE can be connected to one or more CB devices according to the network requirements. No other connections can be established between the PE and the PE. In terms of model, PE is equivalent to a CB remote interface board. Functionally, the vertical Fabric connection between CB and PE is equivalent to the "backplane" of the frame device ". From the management point of view, all CB and PE devices form a stack, which is a device and a management point.

The entire topology has two aspects: one is that multiple CB devices establish horizontal stacking Based on IRF rules and Topology computing; the other is that CB sends HELLO messages to the outside through the longitudinal Fabric port, establish a vertical Fabric based on PE feedback.

As shown in 3, the vertical Fabric creation process is divided into four steps:

Step 1: complete the assignment and acquisition of the expansion board number Slot-ID. When the VCF Fabric port is enabled on CB, the probe packets are sent cyclically. Once the Slot-ID allocation is complete, the probe packets are stopped.

Step 2: complete software loading. It includes several sub-processes, such as sending a load request by PE, providing the description of the version file by CB, and confirming that the load is complete. In this case, Bootware is similar to the BIOS on a PC) and the App is the host software.

Step 3: restart PE with the downloaded version and complete the registration at CB.

Finally, CB sends configuration information to PE.

Figure 3. PE addition and VCF creation process

2. VCF Fabric connection method

As mentioned above, the vertical Fabric connection between PE and CB is similar to the "backplane" of the frame device. In order to increase the bandwidth and maintain a suitable convergence ratio for upstream and downstream traffic, the links between the two are generally composed of multiple physical lines, which can be implemented logically using the HASH method. If a link is Down, the downstream port of the server will not be Down, but the bandwidth will decrease, and the related traffic will be re-hashed and allocated to 4 of the remaining link ).

Figure 4.VCF Fabric connection mode

3. PE Management

Horizontal configuration, Master election, and overall stack creation and maintenance are exactly the same as before IRF has no vertical functions. After Vertical VCF and PE are added, the creation process is relatively complicated. However, essentially, all CB and PE form a single logical entity. You can use any user management interface on CB, such as Console port, Telnet port, or Network Management port for configuration and management.

The system uses the Member ID (Member-ID) to identify and manage Member devices. In an IRF, the Member numbers of all devices are unique. The member number is introduced into the port number to facilitate user configuration and identification of interfaces on member devices. Similarly, in The VCF vertical), the system uses the expansion board number Slot-ID to identify and manage the vertical expansion device, the expansion board number is unique throughout the system and is also introduced into the port number. If CB is a frame device, the serial number must not be the same as the serial number of the existing interface board LPU on the frame device. The Mechanism is slightly different in terms of use. The Member number Member-ID must be restarted to take effect. The expansion board number Slot-ID can take effect immediately after being configured on CB.

Add PE. When a new member device is added to the VCF system, the process varies depending on the system status or the status of the PE device. Assume that the horizontal IRF has been configured and the Slot-ID has been assigned to the PE on CB. 1) PE can be plug-and-play with the default factory configuration. If a longitudinal VCF system runs normally, it is automatically restored if it is powered off or restarted due to some external factors and the system does not need to intervene. 2) during the operation, the PE can access the system at any time through the longitudinal Fabric port. CB automatically calculates the topology to prevent loop generation during the new PE access. From the virtualization point of view, this process is equivalent to inserting the interface board of the frame device. Of course, because the "backplane" link is a dynamic port at this time, topology computing is required to block the loop. The actual frame device has completed this action during initialization.

PE left. PE exit is relatively simple. When the CB and PE connection cables are pulled out or the corresponding port is Down, the system generates a remote interface board exit event. This process is basically consistent with the interface board pulling of the frame device.

4. As CB

When a box-type device acts as a CB and goes down to the PE, the virtual device formed by horizontal CB through IRF is equivalent to a frame-Type Distributed device Main Control Board; vertical PE is interconnected through VCF to form a distributed device interface board or line card for Virtual Frame devices ). The horizontal IRF interconnection cable simulates the interconnection of the Main Control Board in the switching backplane. The Master in the IRF is equivalent to the Master Control Board of the virtual device, and the Slave is equivalent to the backup Main Control Board. Similarly, the vertical vcf cb and PE interconnect cables to simulate the link between the interface board and the backboard in the switching backboard. The PE device is equivalent to the I/O interface board of the virtual device. 5. The logical view of the virtualization device is displayed on the right.

Figure 5. VCF virtualization device formed when the cartridge device acts as CB

5. Frame device as CB

When a frame-type device acts as a CB and is down-mounted with a PE, the virtual device formed by the box-type device through IRF is also equivalent to a frame-Type Distributed device, in this case, the virtual Frame device has more main control boards and interface boards. For vertical lines, PE forms a distributed interface board of the Virtual Frame device through VCF interconnection. The Master Control Board of the Master in the horizontal IRF is equivalent to the Master Control Board of the virtual device, the Slave and Slave Master control boards of the Master are equivalent to the Slave control boards of Virtual Devices and can act as the interface boards at the same time.) the Master and Slave interfaces continue to act as the interface boards, some or all ports of the interface board are connected to the PE. Similarly, for vertical, the box PE is connected to CB through VCF and is generally a frame CB interface board), the PE device is equivalent to the I/O interface board 6 of the virtual device, the logical view of the virtualization device on the right ).

Figure 6. VCF virtualization device formed when the frame device acts as CB

Ii. VCF System Management

As mentioned above, the entire Fabric system can be managed as a logical entity through an IP address. But at the system level, how should we manage software versions, how should we configure and use plug-and-play to build a VCF system?

Software Version Management. IRF compares versions when creating a horizontal stack, and eventually all members will be consistent with the Master version. For VCF, when a PE is added to a stack, it downloads the version from CB. When CB is an IRF stack, the PE gets the version from the Master regardless of whether it is directly connected to the Master. Therefore, the stack system version is consistent with that of the Master. The version obtained from the horizontal stack Slave is synchronized with the version running on the Master node. The vertical Fabric PE obtains the part suitable for the PE operation. Generally, CB and PE are composed of different CPUs and switching chips. Therefore, there are actually two software packages for different functional purposes on CB or Master, when the system is started or running, it automatically retrieves the required information.

Configuration Management. When the entire Fabric system is managed as a logical entity, it can be configured through IRF members such as the Master or Slave Console. Generally, PE does not provide Console and other configuration ports. For VCF, after specifying the Slot-ID corresponding to the physical port or logical aggregation port on CB, And the PE has been added to the system normally, in this case, you can configure PE through CB, such as VLAN and QoS rules of the port on the PE. After the system saves the configuration, the PE configuration information is saved on CB. When the system restarts or changes the PE, the configuration information of the PE also goes up and down from CB, that is, the PE configuration can be "inherited ".

Plug-and-play PE. PE is equivalent to an interface board of VCF virtualized frame devices. The actual frame devices are plug-and-play through hot swapping. In order to achieve similar functions and simplify management, PE supports plug-and-play by connecting Up/Down events through vertical Fabric ports and vertical Fabric. The PE "insert" virtual box creation process is the same as Figure 3 vertical Fabric creation process, which is not described here.

Iii. VCF upper-Layer Control Protocol

VCF focuses on the I/O port expansion of CB devices. In addition to the functions closely related to ports, other upper-layer protocols are basically implemented on CB. The advantage of doing so is obvious. PE only acts as the interface board to insert a virtual Frame device, which increases the port density and reduces the management network element, and the system controls the management plane to move up, it is conducive to centralized control and network policy management in a large L2 multi-server environment. Second, PE performance specifications are not high, is conducive to cost control.

CB is a 1: N backup model for Master nodes and multiple Salve instances in the horizontal IRF structure. As a management control unit of many vertical PES, CB plays a redundant backup role. Longitudinal Fabric PE is added as an interface board. The Protocol control plane inherits the implementation and advantages of horizontal stacking. For example, the TTL hops of layer-3 packets are added with only 1, and cross-PE aggregation is supported.

Iv. VCF forwarding plane implementation

In general, the CB device in VCF performs better than the PE, and undertakes the business data forwarding decision of VCF, while the PE is mainly responsible for the CB port extender. In a vcf cb device, no matter whether the service traffic comes from a PE device or from a non-vertical Fabric port of the CB device, the table is forwarded based on the purpose of the service message.

VCF unicast forwarding

The uplink direction of VCF is from PE to CB. The traffic from the UNI port of PE is directly redirected to the CB device instead of table-based forwarding. After the CB device receives the service message, it extracts the Extended Port and other information from it, and completes address learning and business control based on the information.

Downstream. If the business packets need to be unicast to a UNI port of the PE, the CB device sends the packets to the PE device through the vertical Fabric interconnection port after the business packet forwarding decision and necessary packet modification. After receiving the Service Message, PE directly extracts the port from it to send the Service Message.

VCF multicast forwarding

The upstream direction, that is, the packet flow from the PE to the CB is consistent with that from the previous unicast. In the downstream direction, for business packets that require multicast and broadcast processing), The CB device copies a business packet for each PE device and sends it to the PE device through the vertical Fabric interconnection port. After receiving such business packets, PE broadcasts the business packets in the corresponding VLAN. If the packets are multicast packets, then, copy the corresponding UNI port list based on the multicast index and send service packets.

VCF multi-link Routing Mechanism

Generally, multiple interconnection links are configured between the CB device and the PE, and when the IRF is stacked horizontally as the CB device, multiple interconnection links can be distributed across different IRF member devices.

The uplink direction from the PE to the CB device. The unicast and multicast modes are consistent. The HASH method is used to make the traffic distribution on the interconnection link more even.

Downlink routing from a CB device to a PE device. For unicast, the shortest path principle is adopted, that is, if the CB device is IRF and the interconnection link to a PE is distributed across multiple IRF members, then, IRF selects the shortest path to the PE. If a single IRF member has multiple interconnection links with a PE, the IRF member performs aggregation HASH routing. The starting point of this principle is to minimize the bandwidth usage of the IRF stack link. For multicast, only one interconnection link is selected to send one multicast copy. When there are multiple users in the same PE, the actual replication is performed in the PE.

Of course, in order to improve the system's forwarding performance and reduce latency, some PES can also provide the local traffic forwarding function.

V. VCF architecture features

Multi-level redundancy and high reliability in VCF deployment

VCF supports multiple CB and PE devices. For CB, both the frame type and the box type support cross-PE redundancy access of servers. In particular, frame CB, because CB uses IRF horizontal stack networking, naturally supports cross-frame and cross-board aggregation, thus providing a wide range of choices for network Redundancy Design. At the same time, the VCF solution not only supports the redundancy of the virtual access layer, but also supports the redundancy of the core aggregation layer, which can comprehensively improve the system-level reliability.

High scalability of VCF Technology

In VCF, CB can be created through horizontal IRF stack and can be composed of multiple devices. For example, H3C frame-type high-end devices can build an IRF stack consisting of up to four devices. This is also true when these devices act as CB devices. A box-type device can build larger IRF and VCF systems.

Different combinations of CB roles in VCF provide greater flexibility and better scalability. This advantage allows enterprises or cloud service operators to smoothly expand their IT facilities based on their business growth.

L2/L3 traffic line rate forwarding

All devices in VCF that assume the CB role, including frame-type devices and box-type devices, support line rate forwarding of layer-2 and layer-3 traffic. No additional boards are required, and L2/L3 traffic is at full line speed.

PE equipment supports dual mode and protects user Investment

H3C PE devices support two operating modes: Standard switching mode and PE mode. You can switch between the two modes through the command line or network management. The device factory is set to the standard switch mode by default. when the device is connected to a device that supports the vertical management of VCF and the vertical feature is enabled, the device can automatically switch to the PE mode, that is, plug-and-play is supported.

The dual-mode feature allows you to select based on your network system construction network needs, without sacrificing the vertical Equipment "plug-and-play" and other simplified management functions, this effectively protects users' investment.

Vi. Summary

IT infrastructure virtualization is both a hot spot and a trend in the future. VCF technology provides a way of thinking for network virtualization and data center server virtualization. VCF vertical expansion technology will help you build large-scale virtual networks and simplify management.