What is our primary use case?
The version we're using depends on the model. We use QFX a lot, including versions 5100, 5110, and 5120. For the 4300 and 4400 campus switches, we're running version 21.4.
The solution is deployed on-premises. Juniper doesn't support cloud.
Juniper just bought an orchestration management platform called Abstract.
Abstract is all greenfield, meaning that it's designed to be built from the ground up. You can't incorporate it with the existing infrastructure.
We have over a million users. I work for AWS, so we do a lot of deployment in Juniper.
How has it helped my organization?
If you know Juniper syntax, it's simple to deploy, simple to set up, very cheap, and cost effective for an average network.
What is most valuable?
Juniper is really good when it comes to the MX router, like edge devices for service providers. The hardware is solid, and a lot of ISPs use MX routers. The SRX firewall is pretty good.
The campus switches and access switches are the most valuable and simple to deploy for Virtual Chassis. Other than that, I don't recommend QFX10000 or 5000.
What needs improvement?
If we're building a data center and have Juniper, we have hundreds of Juniper switches in a leaf and spine design architecture, and it's horrible. Sometimes after an upgrade, half of the even port can't be detected. They don't recognize third-party SFPs, and now they claim that they don't support that anymore. There are a lot of issues with the hardware and software.
Juniper needs to get their code up to par because it isn't stable enough for a leaf and spine design.
If you have over 10,000 VMs or Mac addresses, it doesn't scale.
I would love Juniper to have more stable features, hardware, and code. We have so many tickets open with Juniper. The switch reboot doesn't detect the port.
About 80% of my problems aren't engineering problems, but are Juniper's problems. In comparison, Arista gives me 10% of the problems that Juniper gives me when it comes to hardware and software.
It's falling behind compared to Palo Alto because of its features. Juniper's campus switches and data center switches for the leaf and spine design for the overlay and underlay is horrible. I wouldn't recommend that anyone use Juniper fabric, because there are so many problems, issues, and features that aren't supported.
For how long have I used the solution?
I have used Juniper ethernet switches for 20 years.
What do I think about the stability of the solution?
I would rate the stability as five out of ten.
What do I think about the scalability of the solution?
I would rate the scalability as five out of ten.
If you exceed 10,000 or 15,000 Mac addresses, it becomes unscalable and the network becomes very slow. There are a lot of figures within the data center that Juniper hasn't come up to speed. Now, they claim that they don't support CRB anymore. They support symmetrical IRB, and they support dynamic BGP.
How are customer service and support?
I would rate technical support as five out of five.
How was the initial setup?
What's my experience with pricing, setup cost, and licensing?
For the QFX5000 Juniper switches, the price is about $10,000, but they would probably sell it for $1,000. Nobody can beat Juniper's pricing when it comes to switches. Juniper out-prices everybody, but they get you on the back end, maintenance, service port, and service contract.
It's $3-5 million a year just in service contracts, support maintenance, and service agreements to support Juniper features for thousands of switches. I don't see a good ROI on Juniper because a lot of time is spent on troubleshooting. With Arista, you buy $50,000 switches, but they're stable. They're pretty solid when it comes to the code and hardware.
Which other solutions did I evaluate?
I evaluated Cisco ACI and DNA.
I also evaluated Arista, but a certain company didn't want to go with it because they didn't have the budget. Arista isn't cost effective because it's very expensive.
What other advice do I have?
I would rate Juniper as six out of ten.
It's not a bad product, but there are a lot of bad features, code, and hardware. Juniper does ethernet switching, but they also do campus switching and access switching. Those are very basic setups. They also do next generation data centers. You're supporting these compute nodes, storage, and DNS. The next generation data center fabric is horrible. I built a leaf and spine design in Juniper, but we can't get the MC-LAG to work properly, so we have to use ESI-LAG.
We can't use type five EVPN. We have to use type two EVPN. When you're dealing with internal routing like IBGP, it's very hard to troubleshoot. You can't use features like ASPAP, local preference, and other things like route maps and prefixes. You can't really do any of that because it's all internal.
I wouldn't recommend Juniper. If you want to implement it, then evaluate what you're trying to do. If you're doing a lot of campus switches and access switches, then it's fine. If you're doing a lot of policy-based, zero trust, and micro segmentation, and all these parameters are being driven by security, my advice is not to use it.
Many companies, especially trading companies, stock market companies, hedge funds, and banking companies all use Arista for their lower EC platforms, high speed, and high throughput. I build a lot of major corporations, so I see a lot of companies that do that.
Which deployment model are you using for this solution?
On-premises
Disclosure: My company does not have a business relationship with this vendor other than being a customer.