It is not given that the SDN is going to be successful, though there are very high chances of it due to push from Network operators. Unlike earlier efforts, some vendors have resigned to the fact that SDN will become successful and have/are-having products ready in the market.
Let us see who would be the winners if SDN picks up.
Certainly Network Operators/Service Providers/Data Center Operators are clear winners :
It is logical step for VMWare in the sense that they already have very good Management Software to control virtual machines and Virtual Switches within the servers. Extending this to cover physical switches is logical step.
Merchant L2/L3 Switch Silicon vendors are winners too: Due to open standards (For example: Openflow), the value of proprietary silicon chips (by network device vendors today) diminishes. Merchant silicon vendors would provide all the features with lesser cost due to wider customer base. It not only helps Merchant L2/L3 switch silicon vendors, but also Network operators due to reduces cost of goods.
Few of them or all of Broadcom, Marvell and Intel (Fulcrum) would be the winners if they start adopting SDN standards more aggressively.
Network Device Vendors : (Some will win and some may lose)
SaaS Providers are also winners: Though it is going to take a while, I believe when SDN percolates to Enterprises, Enterprise Networking Administrator departments may like to go for Cloud based Controller solution. Since performance and capacity is one important aspect, Cloud providers might offload some controller functionality to a special appliance which would be deployed in Enterprise networks. Even though it is deployed in Enterprise networks, there is no expectation to control this device directly by Enterprise admins. Admins are still would go to cloud to manage their network. Cloud Servers will internally communicate with the purpose built Slave controller appliances (deployed in Enterprise networks) to do laborious jobs.
Embedded Processor Vendors ( Can keep their value, hence Neutral ): Currently Embedded processors are used to run entire control plane software. One might think that not much processing power is required anymore due to movement of control plane software to controller. On the contrary, there may be more performance requirement on the Embedded processor for following reasons:
Let us see who would be the winners if SDN picks up.
Certainly Network Operators/Service Providers/Data Center Operators are clear winners :
- Since Controller software is expected to provide flexibility for operators to add-on their modules on top of networking operating system and customize it, it allows them
- Gets the overall view of network.
- In case of L2 network, all devices put together can be viewed as one mega switch. Addition and Removal of L2 devices can be controlled from central location without any traffic disruption as flows can be programmable from central location.
- Allows virtualization of the network on logical basis (Example: Tenant).
- Facilities to partition the network with fine granular flow/traffic orchestration.
- New innovations on top of Network Operating system on Controllers.
- Simplify the user interface of their central management systems with more and better control.
- Hierarchical Network visualization and Control.
- Reduce Protocol interoperability testing overhead dramatically as SDN would eliminate many protocols that are normally run among devices under one administration control. Due to this, it also reduces the protocol message overhead on the networks.
- Network Operators have choice of Control Plane software from one single vendor and using network devices from other vendors.
- Network Operators can procure only the relevant control plane protocols as per their network requirements.
It is logical step for VMWare in the sense that they already have very good Management Software to control virtual machines and Virtual Switches within the servers. Extending this to cover physical switches is logical step.
Merchant L2/L3 Switch Silicon vendors are winners too: Due to open standards (For example: Openflow), the value of proprietary silicon chips (by network device vendors today) diminishes. Merchant silicon vendors would provide all the features with lesser cost due to wider customer base. It not only helps Merchant L2/L3 switch silicon vendors, but also Network operators due to reduces cost of goods.
Few of them or all of Broadcom, Marvell and Intel (Fulcrum) would be the winners if they start adopting SDN standards more aggressively.
Network Device Vendors : (Some will win and some may lose)
- With Openflow based SDN, much of the Control plane software is no longer going to be part of the devices as it moves to Controllers. Hence barrier to entry into device market becomes lesser and hence one would see more network device vendors getting into this market. One should not be surprised to see SMB device vendors such as Netgear getting into high end market.
SaaS Providers are also winners: Though it is going to take a while, I believe when SDN percolates to Enterprises, Enterprise Networking Administrator departments may like to go for Cloud based Controller solution. Since performance and capacity is one important aspect, Cloud providers might offload some controller functionality to a special appliance which would be deployed in Enterprise networks. Even though it is deployed in Enterprise networks, there is no expectation to control this device directly by Enterprise admins. Admins are still would go to cloud to manage their network. Cloud Servers will internally communicate with the purpose built Slave controller appliances (deployed in Enterprise networks) to do laborious jobs.
Embedded Processor Vendors ( Can keep their value, hence Neutral ): Currently Embedded processors are used to run entire control plane software. One might think that not much processing power is required anymore due to movement of control plane software to controller. On the contrary, there may be more performance requirement on the Embedded processor for following reasons:
- Need for SSL/TLS TCP connections to talk to Controllers.
- They might need to provide shadow tables to tables in Silicon devices.
- In short term, Merchant L2/L2 switch silicon vendors might not support Openflow in thier chipsets. Rather they may support Cache tables and expect Embedded processors to do the job of creating the cache entries based on the tables populated by Controllers. The traffic to Embedded processors may not be insignificant. Even considering 1% of traffic to embedded processor from L2/L3 silicon, it could lead to 1Gbps in 100Gbps switch. That is not insignificant.
- Some basic services are expected to be supported in Embedded processors such as
- Netflow exporting.
- BFD (Bidirectional Flow Detection).
- Proxy ARP
- As I mentioned in my previous post, some services such as IDS/IPS or Intelligent Classification (Application Detection) might be provisioned from the Controller by uploading virtual appliance onto the Embedded processor. It should have enough processing power to handle this kind of processing. It may even require JVM which means more processing power.
No comments:
Post a Comment