This has been a great discussion, with quite varied viewpoints.
Here are some additional observations on Artisa:
• Here’s how they describe themselves (https://seekingalpha.com/filing/4008317 )
Arista Networks, Inc. is a supplier of cloud networking solutions that use software innovations to address the needs of large-scale Internet companies, cloud service providers and next-generation enterprise. (emphasis added)
• Arista’s approach to the “Campus” market is a long term play, involving AI. They call it “Cognitive Campus Campus.” Some may have read the announcement as hardware (the 7300X3 can support 256 100-Gig ports), but it’s really software, actually AI software. There are a few pieces here:
First, whereas most systems today store only a sampling of what’s going on (telemetry data), Arista stores it all in their SysDB (https://www.arista.com/en/solutions/telemetry-analytics ). Sampling means you’re only taking state every 10-15 minutes, which is not only a lifetime in networking behaviors (Heck, even the 5-second intervals some vendors have isn’t good enough to really debug problems). Polling means you’re only getting averages, not actual events as they occur. With Arista, you get every state change, both as it occurs, and saved for later forensics or compliance/auditing.
Second, Arista includes an Analytics Engine that can find trends and problems, and since all the data from all Arista software controlled devices is together, it does so across the whole picture, not just isolated components of sub-sub-networks. In addition, SDKs and APIs (part of NetDB) enable other data analysis tools to be applied. (https://www.arista.com/en/solutions/telemetry-analytics )
But, back to AI. Arista’s CEO, Jayshree Ullal, has a blog that is almost always interesting and technical. She actually telegraphed Artisa’s interest in AI back in Feb: https://blogs.arista.com/blog/-the-future-of-cognitive-cloud…
AI servers together with an Arista leaf-and-spine network and storage appliances can form an important AI nucleus. We have tested these solutions with NVIDIA and Pure Storage to offer the highest IO density per appliance. …
In small steps, Arista has already begun its journey through CloudVision’s® machine learning implementations. If there is an abnormal traffic rate, anomalies are quickly pinpointed and corrected.
I bolded that last paragraph. This isn’t just about fast ports - that kind of speed thing went out with Juniper decades ago. It isn’t even about the configurability of SDN anymore (although Arista provides automation there as well). It’s about things like self-healing networks.
There are 2 aspects to Arista’s approach:
- The “collapsed” Spine topology that gets rid of tiers and more and more reduces the differences between routers and switches. This means instead of a fixed hierarchy of sub-networks that restricts how data can flow, you can literally place everything on a single tier and then use software to configure how traffic flows.
- A software-driven, AI-assisted, management process that not only lets you know what’s going on, it makes it easy to configure and re-configure networks, perform updates across multiple devices with a single click of the mouse, and now even fixes things for you.
I’d add that the security enablement aspects of SDN should not be overlooked. Firewalls are so 20th century. IoT is become more widespread than ever before, which means orders of magnitude more devices are sending and receiving data over networks than ever before.
• Arista has always been good about helping customers migrate to their solutions. Networking customers don’t want upheavals and can’t tolerate long down times. They are partners with companies like VMWare, Aruba, etc.
This is a well-thought out article on where Arista is heading: http://www.eweek.com/networking/arista-introduces-new-switch…
It talks about Ullal’s subtle attack on Cisco’s “intent-based” approach, and describes that breaking into the campus market will not be an overnight affair:
The move into the campus opens up a host of new possibilities for Arista, according to Brandon Butler, senior research analyst with IDC. “They’re really saying that there’s no reason to manage your campus architecture differently than what you’re doing in the data center,” Butler told eWEEK.
Not all customers will gravitate toward such a unified approach to managing data center and campus networks, but those with heterogeneous campus environments and those looking to refresh their architectures could see benefits.
However, there will be challenges for Arista, the analyst said. The campus network market is fairly well-established, and it will take time for Arista to build up the salesforce and go-to-market strategy to make a significant run.
I know the trading style emphasized on this board is like the Pony Express (https://www.history.com/news/10-things-you-may-not-know-abou… ), where you ride a fast horse until it tires and then switch to a fresher horse. However, I still think there is a place in most portfolios for long term buy and hold, for picking a good steady ride that doesn’t require weekly or even daily review, yet will provide rich returns over time. I agree that it remains to be seen whether ANET is truly that type of company, but it looks promising to me still.