Cisco Subnet An independent Cisco community View more

Sorting out the Juniper data center switching puzzle

Answering the remaining questions following Juniper's release of the EX9200 data center switch.

Last week, the esteemed Jim Duffy wrote this news article covering the release of the Juniper EX9200 data center switch. The article left many questions in my mind, including when should customers buy an 8200 now? What about QFabric? What types of customers are right for the beefiness of a product like the 9200? I had a chance to talk to Juniper about the product and I thought I’d clear up some questions that were still outstanding.

FIRST LOOK: Juniper's core SDN switch

The first notable point with the EX 9200 is that it’s actually based on the MX router, not the EX 8200 switch line, meaning it’s an EX by name only. This seemed a bit odd at first until you understand that the primary use case for the 9200 is as an SDN switch, making programmability a must-have. The EX line was built on merchant silicon, whereas the MX was built on the highly programmable, custom Trio ASIC, giving Juniper the foundation it needs to build a programmable switch. With this product, Juniper hopes to one up its Tasman Ave-based friends over at Cisco. Cisco’s programmability story is centered on the python-based OnePK programming environment. Juniper’s Trio-based 9200 gives the company the ability to program at a chip level, allowing for custom rewrite of the silicon. This should give Juniper some unique capabilities. However, it only does so for one of its three switching lines, whereas Cisco’s OnePK covers many of its products today, including the upcoming new ISRs, allowing Cisco to continue to promote the concept of an “architecture.”

In addition to programmability, the EX9200 brings 40 and 100 Gig-E to Juniper Switching. The 8200 had no 40/100 capabilities and QFabric’s 40 Gig connects are used to connect the Interconnect to the QF Nodes, where 10 Gig-E is the interface used for server connections. The 9200 should be a product that even customers installing today will still have in use 7 to 10 years from now.

On paper, the positioning of the 9200 versus QFabric seems to make sense. The 9200 is meant for customers that want a highly programmable switch as the foundation for a software defined network. QFabric is meant more for customers that want high performance, low latency and a single-tier network. However, the problem with having multiple product lines is what if customers want both? When I was briefed by Juniper, the company said that they can imagine a day, maybe 5-10 years out, when the QFabric type of network is the norm. Well, if that’s true, then why not marry QFabric and Trio together and create a single product line that is programmable with the latency characteristics of QFabric? I think the reason is that bringing Trio on board with existing QFabric would involve significantly more engineering than building a product from scratch.

Regarding the 8200, the company is repositioning the product as primarily a campus switch. I think, given the age of the product and the interfaces on it, that’s the most sensible way to use the product. There’s nothing wrong with the 8200, there are just bigger and faster switches available from Juniper and its competitors. The challenge for Juniper will be what to do with the existing customer base. If you recall, when Cisco launched Nexus, it tried to position Catalyst as a campus switch, which made sense. However, much of the install base of Catalyst did not want to switch platforms, causing Cisco to back off the positioning of Catalyst as campus only. Over the past year or so, Cisco has done the right thing for its customers and beefed up Catalyst so customers that want the fabric benefits of Nexus can buy that platform. If they want to stay with Catalyst, they can do that too. I’m not sure the 8200 is capable of actually doing 100 Gig-E, but regardless, Juniper needs to give its existing 8200 customer base a migration path that is minimally disruptive.

I do believe Juniper needed a next-generation data center switch and the 9200 is a monster of a switch. Like I said, I think the Trio-based product has some unique capabilities with regards to being the foundation for a software-defined network. Trio-based MX has been very popular with its telco customers, and I think enterprise customers will benefit greatly as well. However, I also think there’s some difficulty in carrying multiple data center switching lines forward, especially for a company with the small amount of share that Juniper has. QFabric, EX8200 and EX9200 are all good products. Juniper needs to handle its current 8200 customers carefully and make sure the positioning of one versus the other is crystal clear within its channel if all three product lines are to be successful.

To comment on this article and other Network World content, visit our Facebook page or our Twitter stream.
Must read: Hidden Cause of Slow Internet and how to fix it
Notice to our Readers
We're now using social media to take your comments and feedback. Learn more about this here.