Dan Conde, an analyst with Enterprise Strategy Group Inc., in Milford, Mass., wrote about some current disagreements...
in multicloud strategy.
Many enterprises are now debating whether it's better to work with a single cloud service provider versus two or more public cloud providers. Fans of the one-cloud approach argue that adopting multiple clouds imposes a substantial learning cost, and it makes the most sense to choose one, because many clouds are similar.
On the other hand, multicloud supporters contend that each cloud is different, and companies should endeavor to match multiple cloud providers that best suit different workloads. Like choosing to run both Linux and Windows, these organizations seek to mitigate financial and technical risks.
Conde explored his approach to multicloud strategy. He recommended large enterprises adopt more than one cloud to take advantage of wide-ranging cloud services. However, the best multicloud strategy may be different for smaller companies. These groups may wish to choose a single provider because of limited staff available to learn about the different capabilities of multiple clouds. ESG research indicated that 75% of public cloud infrastructure customers rely on multiple clouds.
Explore more of Conde's thoughts on multicloud strategy.
Why didn't leaf-spine networks emerge sooner?
Ivan Pepelnjak, writing in ipSpace, answered a reader's question about traditional three-tier network architecture. Looking back at the past decade, Pepelnjak said he knew many engineers who constructed data centers around, say, a single Cisco Catalyst 6500, while others used paired Catalyst 6500s with virtual switching system software to behave as a single component.
In other cases, engineers used stackable switches connected to a central pair of Catalyst 6500s. In most cases, these data centers didn't need three-tier network designs, and the forwarding performance of a single box would be sufficient. As a result, Pepelnjak said, some organizations built three-tier data centers that far surpassed their needs, "blindly" following vendor guidelines as they did so.
According to Pepelnjak, data centers with a pair of core switches to connect access switches needed three-tier architecture the most, usually because of weak forwarding performance or port density problems. In these kinds of data centers, bandwidth was often limited by architecture -- the location of endpoints determined throughput.
When it comes down to it, Pepelnjak said top-of-rack switches connected to modular core switches in a multiple link aggregation group cluster fit the definition of a leaf-spine fabric just as much as a data center with many large white box switches. "If you ignore the oversubscription in the core switches, a lot of data centers already have leaf-and-spine fabrics without knowing it," he added.
Dig deeper into Pepelnjak's thoughts on leaf-spine architecture.
Innovium Teralynx joins the switching silicon market
Drew Conry-Murray, writing in Packet Pushers, explored the launch of Teralynx 7, a new programmable switch family from startup Innovium Inc. The new switch silicon offers users 12.8 Tbps of throughput and can support links for 10 Gigabit Ethernet, 25 GbE, 100 GbE, and both 200 GbE and 400 GbE.
The company, based in San Jose, Calif., engineered Teralynx with a 70 MB onboard buffer device for very low-latency instances. The startup is challenging Broadcom, which offers products that allow only limited tinkering with packet processing through proprietary software development kits. Innovium is marketing its product toward very large data center groups.
Another contender, Barefoot Networks, offers an alternative with its Tofino chipset, Conry-Murray said. Yet, Teralynx offers almost double the throughput of Barefoot's switch silicon.
Tofino allows full programmability, while Teralynx has Layer 2 features preloaded on the chip, supporting MAC addresses, VLAN and spanning tree. It also comes equipped with Layer 3 capabilities and support for IPv4, IPv6, virtual routing and forwarding, MPLS and equal cost multipath. In addition, while Barefoot has embraced P4 to enable programmability, Innovium is using the open source development tool for telemetry. The switch silicon is expected to be released commercially in the third quarter of 2017, aimed at top-of-rack and leaf-and-spine switch deployments.
Read more of Conry-Murray's thoughts on Innovium.
Benefits of multicloud
The case for leaf-spine
Corsa expands programmable switches