In this interview, Priya Natarajan, senior director and head of worldwide go to market at Lenovo discusses the evolution of the data center and how artificial intelligence (AI) and machine learning (ML) are being used to streamline workloads and improve reliability.
SDxCentral: Artificial intelligence (AI) and machine learning (ML) make it possible to automate some data center operations, but all this data that is collected can overwhelm networks. How does Lenovo see AI and ML impacting the data center?
Priya Natarajan: Data center technologies are evolving rapidly to keep up with the growing volume of data. With AI and ML coming into full force, the volume of data increases further. By observing patterns within the network, ML can be applied to understand the status of the network, identify bottlenecks and predict how it will react to future workloads. It can then make intelligent recommendations on how to reconfigure the network. It’s even possible to automatically orchestrate the network based on predicted behavior. All this contributes towards better network availability and workload performance and reliability.
The explosion in cloud usage has significantly increased the amount of traffic handled by data centers. How can today’s data centers keep up with these new workloads?
Many enterprises use a hybrid approach to data centers. It is also common to have a multi-cloud approach. Enterprises choose to segregate based on applications and also manage risk by having a multi-cloud approach. As a result, workload migrations happen frequently and rapidly, increasing the amount of traffic handled by the data centers. In order for this to work seamlessly, the data center network must have sufficient bandwidth, be reliable, have low latency and also offer the flexibility to scale with the growing demands and traffic bursts. The networks should be integrated with DevOps tools such as Chef, Puppet, and Ansible to improve manageability and operational efficiency. In addition to that, they should have an intelligent network controller that can automatically identify bottlenecks and redeploy resources to optimize for the workload.
For some customers, obtaining sufficient bandwidth and reliability may include the need to upgrade the network hardware in their facilities. Use of Software-defined Networking (SDN) may also help use network resources more efficiently.
What types of telemetry capabilities are available to help streamline this process?
Capturing telemetry information from every packet at the ASIC level, and analyzing it to extract valuable insights is important to ensure a workload performs as designed. This allows one to verify that the networking indeed behaves as expected and make modifications otherwise. Telemetry capabilities fit into three buckets:
- The first one helps to determine how the network behaved in the past. It is useful to understand what happened in retrospect, understand any past bottlenecks in the network, and what might have caused those.
- Second, it is important to know how the network is handling the workload right now. This requires real-time streaming. This helps with troubleshooting the current workload and optimizing the network.
- Third, it is important to predict how the network will react at a given time to a given workload. This helps to prevent any issues before they happen.
The insights gathered via telemetry can be applied to optimize the network. With the help of SDN and an intelligent network controller, the network can be orchestrated to meet both expected and sudden changes in the workload. Lenovo network telemetry solution alleviates these challenges and provides end-to-end visibility into networks.
Older hierarchical network topologies can create bottlenecks at the network core and hinder traffic flow. Why are newer network topologies like leaf-spine or collapsed-spine a better option? What topologies does Lenovo support and why?
Spine-leaf networks are better architected for cloud and software-defined data centers. With built-in route redundancy, they are optimized for reliability. The architecture of the network determines the overall network performance. In the modern virtualized data center, the majority of the communications happen within the data center. East-west traffic is growing in the data center and it has a significant share of the total traffic. The full mesh topology of spine-leaf architecture results in a flatter network with fixed-switch configurations. This reduces latency and eliminates bottlenecks, thereby improving application performance. Spine-leaf designs typically also have consistent maximum end-to-end latency for east-west traffic between any two servers within the topology. As a result, workloads running over this network will behave more predictably, which is key for high performance computing clusters or high frequency trading applications.
Lenovo supports both Layer 2 and Layer 3 spine-leaf architectures. The Layer 2 topology offers up to 200 Gb/s and is useful for connecting more servers, either as individual NICs or by bonding with MAC address load balancing. The Layer 3 topology is more complex and offers 200 Gb/s to 600 Gb/s spine speeds. It uses dynamic routing protocols to connect all the spine-leaf switches and the core network. Lenovo offers 25G/100G and 10/40G switches for leaf-spine combinations. These provide the best over subscription ratios. These topologies enable connecting thousands of servers with low latency single hop connections between leaf switches. These switches offer the customer the flexibility to grow their workload while maintaining low latency on a reliable network. Spine-leaf designs lend themselves to expansion easily to keep up with customers’ need to add more bandwidth.
How does Lenovo differentiate itself from its competitors in this area? Why should a customer select Lenovo for its data center needs?
Lenovo offers one-stop shop for customers’ data center needs. Customers get their compute, storage, and networking from Lenovo. These components work better together and can be managed with XClarity, a single pane of glass for management of the data center infrastructure. Lenovo networking offers the lowest total cost of ownership in the industry as validated by ACG Research. Lenovo networking is built with cloud in mind and it scales with the data center traffic. It is open and interoperable with other leading vendors, avoiding vendor lock-in as validated by The Tolly Group. Lenovo networking is future-proof – the customers can update their software anytime with all features included. In addition, we include a three-year basic warranty services with our products. Lenovo continues to innovate for the software-defined data center, including offering special and unique productivity features to reduce network updates, support for various cloud ecosystems like Microsoft Azure, VMware, Nutanix, and OpenStack.