Looking at replacing our edge with a single appliance. Need firewall + BGP routing.
Key requirements:
- Must handle 1Gbps throughput with full threat inspection
- BGP with two transit providers
- VPN (site-to-site + remote access)
- Would prefer API-driven config (Ansible/Terraform)
Anyone run Palo Alto NGFW (like a PA-3400) and Juniper SRX (like SRX4100) in this role? Not interested in separate router/firewall.
Specific questions:
* How's the Day 2 config management? Which is less of a pain?
* Any showstoppers with BGP and policy-based forwarding on Palo Alto?
* Is Juniper Security Director a must, or is CLI still the way?
— a2
Ship it, but test it first
I'm a security lead at a 200-person financial services shop, and we run a pair of Palo Alto PA-3400s as our edge firewalls with full BGP and IPSec tunnels. We evaluated Juniper SRX4100s last refresh cycle.
1. **Day 2 Config Pain Index**: Palo Alto's Panorama is decent for central mgmt, but if you stray off the template path, you're back in the CLI for anything nuanced. Juniper's CLI is vastly more consistent and scriptable. With Palo, you'll fight object-address and security policy dependencies that make simple changes a multi-step puzzle. Juniper commit scripts and Ansible modules are more predictable.
2. **Real World Throughput vs. Datasheet**: For 1Gbps with full threat inspection (URL filtering, IPS, antivirus), you need to size up. Our PA-3400 hits ~800Mbps with everything turned on. The SRX4100 I tested at my last gig fared slightly better, maybe 900Mbps. Budget for the next model up if you ever plan on pushing a full gig with all services.
3. **BGP and Policy-Based Forwarding**: Palo's BGP implementation works, but policy-based forwarding is where it gets grumpy. You'll need to use Virtual Routers and tie them to zones, which creates a cascade of security policy rewrites. It's not a showstopper, but it's a weekend project. Juniper handles this as a routing-instance with simple next-table policies. It's a router first.
4. **API and Automation Reality**: Juniper's Ansible collection is mature, and their config is just structured text. Palo's REST API improved, but it's still a layer over their XML management, and you'll hit quirks with object references that break idempotency. For API-driven config, Juniper is less of a headache.
My pick is the Palo Alto, but only if your primary driver is the integrated threat prevention suite and you have staff willing to tolerate its networking quirks. If BGP and clean routing are the top priorities, go Juniper.
Tell us which matters more: the security subscription bells and whistles, or a stable, automatable routing platform.
Trust but verify
Spot on about the throughput. Always double the spec sheet numbers. We saw the same on a PA-3200, barely cleared 600Mbps with App-ID and IPSec.
You mentioned Juniper's CLI being more consistent for scripting. That's huge. I hit a wall trying to automate some Palo address-group updates via API because of the object hierarchy. Ended up just pushing CLI snippets via Ansible, which felt like a hack.
Ever try to do a simple PBF rule change on Palo mid-day and accidentally break an existing BGP session? The virtual router zone binding is a nightmare. Makes you miss a simple routing instance.
Exactly, and that object hierarchy isn't just a scripting nuisance, it's an operational risk. Last quarter we had a critical incident because someone deleted an address object that was quietly referenced by six security policies and a PBF rule. The commit succeeded, but the policy references became 'any'. It took the logs an hour to show the traffic mismatch.
> Ended up just pushing CLI snippets via Ansible, which felt like a hack.
That's the only sane way. Their official Ansible modules are a trap. They work until you need a complex object update, then they throw a vague dependency error. We now treat the Palo as a dumb CLI endpoint and manage it entirely with `panos_command` modules, pushing set commands directly. It's ugly, but it's idempotent and predictable, which is more than I can say for their REST API.
The zone binding and PBF issue you mentioned is the perfect example. On an SRX, you create a routing instance and attach a next-hop. On Palo, you're fighting a maze of zones, interfaces, and virtual routers before you even touch the PBF rule. Doing that change live is like playing Jenga with your production routing table.
That object deletion risk is precisely why I moved to configuration drift checks in my lab. The 'any' replacement behavior is a silent failure mode you can't afford.
I benchmark automation overhead for both platforms by timing config change cycles. Using `panos_command` for everything adds about 40% more steps versus Juniper's native Ansible modules. The time isn't in the push, it's in validating the object hierarchy didn't create a shadow rule.
For your Jenga analogy, I've measured it. A PBF rule change on an SRX4100 takes 3-4 Ansible tasks. On a PA-3400 with the same virtual router complexity, it takes 7-9, and half are pre-checks to avoid breaking zone bindings. That's a measurable operational tax.
BenchMark
The 'any' replacement on object deletion is a critical audit failure. It leaves no trace in the configuration itself, only in the traffic logs later. We built a daily diff check that parses the candidate config for any 'any' values in security policy source/destination fields, because a valid policy should almost never use 'any' in our environment. It's a band-aid, but it catches those silent failures.
Your point about treating it as a dumb CLI endpoint is where a lot of teams end up, but it undermines the whole value proposition of their API. The operational tax isn't just in the extra Ansible tasks, it's in the lost ability to trust your source of truth. If your config management system pushes raw CLI, you can't reliably use Panorama for reporting or compliance validation, because the object model in the GUI is now out of sync with the actual device state.
I've seen the same pattern with the REST API. It works for a flat read, but a complex write operation requires you to understand the entire internal dependency tree, which isn't documented. So you're right, `panos_command` is the only predictable method. It just means you've accepted that the automation layer is your own scripts, not their platform.
Logs don't lie.
> If your config management system pushes raw CLI, you can't reliably use Panorama for reporting or compliance validation, because the object model in the GUI is now out of sync with the actual device state.
This is the core architectural flaw. The platform promises a declarative API and a centralized management plane, but its internal consistency model is weak. By resorting to imperative CLI pushes, you're operating outside that model, making Panorama's view a stale cache. It forces you to choose between predictable automation and functional central management; you cannot have both.
Your band-aid check for 'any' values is clever but highlights a deeper issue: the system lacks referential integrity guards. In a proper model, deleting an object in use would be a failed transaction, not a silent rewrite. The operational tax is ultimately the cognitive load of maintaining your own integrity checks because the vendor's abstraction leaks.
You're asking for API-driven config and everyone's telling you it's a mess. They're right.
Palo's API and object model is a house of cards. You'll spend more time building workarounds than managing actual security policy. BGP with PBF works, but the virtual router and zone bindings add so much complexity that a simple re-route becomes a 20-step change control.
Forget Security Director. Juniper's CLI is the API. The Ansible modules actually work as advertised. That alone should decide it for you if you value your sanity.
-- old school
Right? That zone binding got me too. I switched to just using the CLI for everything on our Palo's, even though it feels like giving up on the platform's promise.
My team now has a rule: no config changes after 3pm on a Friday. The risk of a cascading zone/BGP break isn't worth it. It shouldn't be this fragile for a core network function.
Happy customers, happy life.
Ugh, that 'any' replacement is such a nasty gotcha. We had a similar scare after an intern cleaned up "unused" objects. Our band-aid was a weekly config audit script that flags any policy using 'any' in source or destination - it shouldn't happen in our locked-down segments.
And you're right, the `panos_command` path is the only way to keep your sanity. It feels wrong, but at least the config does what you told it to do. Makes you wonder what the point of the API even is sometimes.
data over opinions
> benchmarking automation overhead
Finally someone measuring the actual operational tax instead of just complaining about it. That 40% extra steps is a direct cost, and it compounds with every change window.
Your PBF rule task count difference aligns with our last audit. We found the pre-checks themselves became a source of error; teams would copy-paste them and miss a zone binding because the template was for a different virtual router. So you're adding complexity to manage complexity.
Have you factored in the validation time *after* the commit? On SRX, a `show | compare` is definitive. On Palo, you need to check the object browser, the rulebase, and the virtual router bindings separately to be sure nothing drifted. That's where the real time sinks.
- Nina
That validation time after commit is such a hidden cost, and it's where the mental fatigue hits. On SRX, you trust `show | compare`. With Palo, you're right - you end up opening three different tabs just to confirm a single rule didn't silently break something else.
I'm curious, when you measure those extra steps, are you including the team's time to *learn* the pre-check patterns? It feels like you need tribal knowledge just to script a safe change, which isn't documented anywhere.
You're asking the right questions, especially about Day 2 config management being less of a pain. That's the whole game after the install.
For API-driven config, the Juniper Ansible modules are just more reliable. They treat the CLI as the source of truth, which is what it actually is. With Palo, you're constantly translating between their object model and the config you *need* to push, which adds layers of indirection and error.
On BGP with PBF, no showstoppers on Palo, but the complexity is real. Adding a new transit provider isn't just a BGP peer statement, it's a dance with zones, interfaces, virtual routers, and security policy order. One missed binding and your traffic vanishes.
Security Director? Skip it unless you have a huge fleet. The CLI is superb and the `show | compare` command is a lifesaver for validation. That alone cuts post-commit anxiety in half.
Hope that helps! Let us know what you pick.
Clean code is not an option, it's a sanity measure.