Network solutions for enterprise data centers are evolving to support better agility and cloud architectures. Enterprise I&O leaders should evaluate multiple vendors, and focus on comparing architectures, software features and infrastructure integration capabilities, not hardware specifications.
The data center networking solutions covered in this research are hardware and/or software solutions to deliver connectivity within enterprise data centers. Network technology solutions include data center core/spine switches, server access switches (top of rack [ToR], leaf), virtual switching, programmable Ethernet fabrics, network automation, network overlays and emerging approaches, based on disaggregation of hardware from software and open-source components.
The data center network historically was just a fast and scalable LAN to connect all data center equipment. Today, with the increased adoption of cloud computing models and in support of digital business initiatives, enterprises need better integration and orchestration of the network with the rest of the data center infrastructure (storage and compute). In addition to the highest levels of performance and high-speed Ethernet ports (10/25/50/100GbE), data center network equipment must support automation, programmability and analytics for application performance monitoring and troubleshooting.
We envision that functionality for building a hybrid cloud, through seamless integration of enterprise-owned network solutions with public cloud services, will play an increasing role in vendor selection by 2020.
This research evaluates data center switching solutions for enterprises that procure and manage their own data center infrastructure, for installation within their premises or in colocation facilities. While public cloud providers and hyperscale data centers are also large consumers of data center switching solutions, their buying criteria and requirements significantly differ and are not considered in this research. The data center LAN market is now clearly differentiated from campus LANs, which include wired and wireless access infrastructure, covered in “Magic Quadrant for the Wired and Wireless LAN Access Infrastructure.”
The enterprise data center networking market in 2016 grew by 13% in revenue (see “Market Share: Enterprise Network Equipment by Market Segment, Worldwide, 4Q16 and 2016” ). 10GbE was the most common connection for servers, and the average cost of a 10GbE port declined by 22% (from $214 to $165). The 40GbE segment experienced strong revenue growth (71%) in 2016, but we expect this growth to shift to 50GbE and 100GbE in 2017, which will quickly replace 40GbE for new purchases.
This market is mostly driven by:
- Replacement of existing data center switches that are becoming obsolete
- Expansion of network infrastructure to support new applications/services
- Adoption of new solutions to increase agility, ease of operations and cost-effectiveness of existing infrastructure
Simultaneously, this market is being impacted as some organizations desire to shrink their on-site infrastructure, by moving variable portions of their workloads to colocation facilities or public cloud.
Along these lines, the common topics that data center networking clients have raised in the past year include:
- What is the status of software-defined networking (SDN)? Should I consider SDN?
- What commercial solutions (including programmable fabrics and overlays) can increase automation and agility? (Most questions were regarding Cisco ACI and VMware NSX.)
- What vendors are best-aligned with future requirements?
- How do the different commercial solutions compare?
- What network solutions will facilitate the implementation of a hybrid cloud?
- Is my proposed purchase competitive from a pricing perspective?
- Are hardware/software disaggregation and open source relevant for the enterprise?
Clients focus on pragmatic deployments of automated fabrics, with purchases usually triggered by the need to replace obsolete equipment and deploy 10GbE (or, increasingly, 25GbE) for server attachment. They are interested in basic fabric solutions (SDN-based or traditional), including virtual network overlays that can be deployed incrementally when a hardware refresh is not planned.
Data center networking equipment is business-critical and has a long life cycle (typically four to seven years), so buyers are strongly influenced by historical vendor relationships, experiences with the quality of their support and technical familiarity with previously installed products. Uptake of more innovative solutions is relatively slow, with risk adversity, skill sets, and cultural or organizational barriers being the biggest inhibitors.
Although the need to contain cost is frequently mentioned, price is not the main driver for most enterprise buyers, for both equipment and support, and the availability of lower-cost alternatives is often not enough to trigger a supplier change.
What is driving change in data center infrastructures is the need for improved operational agility, a key requirement as clients go through the digital transformation of their businesses. Vendors are evolving the features of their solutions in response to this pressure, but the biggest changes since the previous edition of this Magic Quadrant are related to the vendor landscape. It is essential for clients to understand the strategic market position of all vendors, in addition to the technical attributes of their solutions, to ensure that they do not put themselves on a dead end.
The landscape of data center networking vendors is rapidly changing, with consolidation taking place among smaller players and changes in strategy for leading vendors:
- Broadcom announced the intent to acquire Brocade, which has a cascading effect on this market, as Broadcom is also a supplier of chipsets to most vendors and immediately announced its intention to divest the IP Networking division.
- The IP Networking division of Brocade is being split and sold in separate parts, with Extreme Networks announcing the intent to acquire Brocade’s data center switching, routing, automation and analytics business.
- Avaya filed for Chapter 11 bankruptcy protection, and as part of the restructuring, Extreme Networks made a stalking-horse bid to acquire the data networking portion of Avaya.
- In 2016, Hewlett Packard Enterprise (HPE) completed the sale of 51% of its H3C subsidiary in China and this had major implications for HPE’s strategies in the data center networking market. In November 2016, HPE announced a partnership with Arista Networks and a strategy to offer its customers and partners the ability to purchase Arista’s switching products directly from HPE. Thus, Gartner sees HPE’s long-term role in this market to be primarily a system integrator and reseller, versus a direct competitor to other vendors.
- In 2016, Dell completed the acquisition of EMC, which included VMware. Dell officially stated that VMware will continue to operate completely independently. We have not seen any effect of the acquisition on the strategy and the networking portfolio of both Dell and VMware so far, but as the joint company makes progress with operations in 2017, there might be synergies or strategy adjustments.
Other emerging technological and financial trends include:
- 25GbE switch ports starting to replace 10GbE for server access at similar price points
- 50/100GbE interswitch connections replacing 40GbE as a higher performance, efficient and cost-effective alternative
- Brite box and disaggregation of hardware and software gaining some traction, as demonstrated by the inclusion in this Magic Quadrant of new vendors focused on software that meet inclusion criteria for the first time (such as Big Switch Networks and Cumulus Networks)
- Streaming telemetry and application analytics becoming a key selling point for new data center networking solutions, because they provide insight that facilitates operations and will enable intent-based networking
- Flexible pricing models, including pay as you go, subscription service pricing and burst capacity pricing being available, although we have not yet witnessed significant adoption in enterprise organizations
- A split of hardware and software pricing in proposals, both for equipment and support (even when hardware and software must be purchased from the same vendor, such as with Cisco One; see “Verify If Cisco One Will Save You Money Before Adopting It” ); this signals a vendors’ mindset change and indicates that the disaggregation trend will continue
- An increasing use of Internet Protocol (IP)-based storage networking, which drives more traffic to the data center network, instead of to the traditional Fibre Channel SAN
- Deployment of hyperconverged integrated systems (HCISs) from vendors like Nutanix or HPE SimpliVity creates a need for their integration with network and security architectures, beyond pure connectivity
- Adoption of private, public and hybrid cloud models in enterprises demanding support of network capabilities like multitenancy, segmentation, policy management and end-to-end monitoring, with solutions that can consistently span private, public and hybrid environments
- Networking solutions for containers starting to appear, from the vendors covered in this research or from open-source initiatives; they will be needed as containers migrate from their current use in development environments for Mode 2 applications to production deployments over the next two years
There is an ongoing move away from using device-by-device command line interface (CLI)-driven configurations toward a centralized and more policy-based mode of operations. We expect that, by 2020, only 30% of network operations teams will use the CLI as their primary interface, down from 85% at YE16 (see “Predicts 2017: Enterprise Networks and Network Services” ).
Today, most data center networking solutions come with a GUI, so adding network configurations can be done through a simpler interface and the necessary changes are then automatically applied to multiple network devices. However, the real innovation comes with the API, because it enables complete automation of repetitive tasks, but also integration with higher-level infrastructure automation tools that can leverage the network API directly, without involvement of a human network operator, at least for a subset of tasks (such as provisioning of network services to newly created virtual machines [VMs]). APIs and device-level automation tools (like Ansible, Puppet or Chef) can also be used to implement a DevOps model.
While there is consensus on the increasing importance of software, there are still vendors investing in proprietary hardware (like Cisco Cloud Scale and Juniper Q5 application-specific integrated circuit [ASICs]), because they believe that developing the entire system gives them a better chance to lead in some aspects and differentiate from competitors, at least in specific use cases.
That said, most vendors are now focusing their innovation efforts on software and leveraging merchant silicon to build their switching portfolio. Many vendors are using more than one chip supplier, and we are seeing increased competition in the merchant silicon market, which offers OEMs the ability to optimize platforms based on functionality and price. This creates a need for making the software adaptable to different chipsets. While many vendors use the software drivers provided by the chipset manufacturer to simplify the task (leveraging open platforms like ONIE and SAI), others prefer to develop the entire stack to achieve better optimization and differentiate. Some chipsets can be dynamically configured for different switch usage scenarios (for example, a large MAC table versus a large routing table) or provide highly programmable frame processing (for example, with languages like P4), so users will benefit from more options and choice as vendors differentiate their solutions through software, even when using the same chipset.
Given the broad spectrum of software options, we believe that merchant-based platforms can meet the needs of at least 80% of enterprises.
Most vendors offer fabric solutions that are based on different (and generally proprietary) architectures, but can provide similar benefits as SDN, at least in terms of programmability and centralized point of control, if not in terms of reduced vendor lock-in and cost. Examples are Cisco ACI and NFM, Arista Networks CloudVision, Juniper Networks Junos Fusion, and Huawei Cloud Fabric.
All vendors are putting a great emphasis on network analytics and delivering solutions that provide increasing insight into the application data flows that are traversing the data center network. Streaming network telemetry data is a new mechanism to feed analytical tools, and scales well beyond the traditional SNMP polling of Management Information Bases (MIBs). These capabilities enable easier troubleshooting of traditional network issues, like identifying congested network interfaces that are dropping packets, but they give also infrastructure and operations (I&O) leaders a lot more information about applications and how the network is influencing the service level experienced by end users (for example, Cisco Tetration Analytics and Arista DANZ).
Analytics combined with automation capabilities are foundational to intent-based networking, a concept that represents the next frontier for vendors. With intent-based networking, high-level business policies are translated into network configuration commands and implemented through automation. Real network behavior is then compared against the desired behavior to enable any necessary remediation and ensure that the policy is actually enforced. The idea is to build a self-driving network. Just like building a self-driving car, implementation of intent-based networking systems requires data from sensors (network analytics) to feed the computation of abstract models built with complex artificial intelligence algorithms, and then actuators to operate the car (network automation through the API; see “Innovation Insight: Intent-Based Networking Systems” ).
There is a growing willingness to move away from proprietary solutions. Our end-user survey (n = 83) indicated that 42% of clients consider open standards and multivendor interoperability support a mandatory requirement, 34% consider it very important, and 20% consider it somewhat important, so openness is a relevant buying criterion for 96% of the end users.
These results are quite impressive, but in addition, 75% of the end users indicated that they expect an increase in relevance of open networking in their purchasing decisions in the next 24 months. This requirement is mostly unfulfilled by vendors, since the majority of the solutions considered in this research are proprietary; thus, it might create market opportunities for players willing to pursue this path.
Disaggregation of hardware and software can be seen as a first step toward more vendor independence. Interest and adoption of white-/brite-box switching has increased significantly within hyperscale data centers over the past several quarters, and we expect it to reach 22% of the total data center Ethernet switch market by 2020 (see “Forecast Overview: Enterprise Network Equipment, Worldwide, 2017 Update” ). Enterprise clients were inspired by hyperscale stories and asked questions on suitability for their use cases. As a consequence, we have seen enterprise adoption of brite box (the most successful delivery model in this market segment) grow, and we estimate about 1,000 paying enterprise customers as of March 2017. This includes some very large accounts in finance and the public sector. However, established vendors (for example, HPE and Dell) did not generally lead in enterprise with solutions based on brite box and third-party software, or did that only for specific use cases like network packet brokers, so adoption of disaggregated solutions in mainstream enterprises is still limited.
HCISs tightly couple compute, network and storage hardware in a system and are gaining popularity, since they simplify operations and improve provisioning times. The networking components of an integrated system are largely prescribed, which results in the transfer of the physical access layer network buying decision from what was solely a networking decision to an integrated server/storage/network decision. Examples are Nutanix and HPE SimpliVity.
Scaling of HCIS clusters and deployment of input/output (I/O)-intensive workloads will lead to the necessity of an end-to-end network design to achieve a more robust, cost-effective and reliable solution (see “Leverage Networking to Ensure Your Hyperconverged Integrated Systems Can Support Demanding Workloads” ).
Containers are a technology for running multiple workloads under the same OS kernel instead of using separate VMs, each with an OS. Container networking is at the maturity stage where VM networking was about five years ago; it is fast-evolving and fragmented, with all vendors in this research starting to provide solutions and launching initiatives. It is difficult for network architects to determine which vendors and architectures are best-suited for their usage scenarios, but at this time, containers have very limited production deployments in enterprise data centers. However, this will be an important decision in the next three years, as containers are widely used for development of Mode 2 applications that sooner or later will need to be deployed in production (see “Take [Limited] Action to Prepare Your Data Center Network for Containers” ).
[to continue, click HERE]