Skip to content
Notifications
Clear all

Complete newbie here - where to start with a Proof of Concept for a 500-user company?

5 Posts
5 Users
0 Reactions
1 Views
(@datadog)
Estimable Member
Joined: 1 week ago
Posts: 90
Topic starter   [#8844]

You're not doing a POC. You're doing a production evaluation. Call it what it is. With 500 users, a misstep will be an incident.

Start with a clear 30-day objective. Define success metrics **before** you rack the hardware or spin up the cloud instance. Your goal is to validate it works for *your* traffic and doesn't break anything.

**Critical POC phases:**
* **Baseline Week:** Document current firewall throughput, session counts, and top application protocols. Use existing tools. No Check Point code yet.
* **Deploy & Mirror:** Deploy in parallel, use passive or mirror ports. Test policy logic against live traffic. This is where you catch bad rules.
* **Controlled Cut-overs:** Start with non-critical segments (guest Wi-Fi, a single server VLAN). Measure latency impact, SSL inspection breakage.
* **Failover Test:** Schedule it. Pull the power. Document the RTO.

**Key metrics to track from day one:**
* CPU/Memory at peak throughput with all blades enabled.
* HTTPS inspection latency (compare to baseline).
* Threat prevention false positives per policy hit.
* Management server stability (this is often the bottleneck).

Avoid the sandbox. Test with real, anonymized traffic or you'll waste everyone's time. Their sales engineer will push for a quick win; hold them to your schedule and metrics.

—DD


Metrics don't lie.


   
Quote
(@integration_maven)
Estimable Member
Joined: 4 months ago
Posts: 130
 

Absolutely agree with calling it a production evaluation. That framing changes the entire stakeholder conversation.

One additional metric I'd track in the baseline and controlled cut-over phases is API traffic, especially for any internal or SaaS tools. A rule that works fine for web browsing can silently break an OAuth flow or a webhook, and users might not report it for days. I'd mirror that traffic and run it through a dev instance of the critical app to catch authentication failures.

Your point about the management server is key. In a 500-user environment, it's tempting to run everything on a single VM, but that's where performance cliffs happen. I'd test pushing policy updates during peak load to see if the management console becomes unresponsive.


IntegrationWizard


   
ReplyQuote
(@alexh99)
Eminent Member
Joined: 1 week ago
Posts: 33
 

Good point on the API traffic. I've seen a Looker dashboard break during a similar eval because of a re-encryption rule. The error logs weren't clear at all.

Is there a reliable way to baseline that API traffic volume, or do you just have to sample it?



   
ReplyQuote
(@consultant_carl_42_v2)
Estimable Member
Joined: 4 months ago
Posts: 115
 

Totally agree on the production evaluation mindset. Framing it as a POC can accidentally lower the guardrails with stakeholders.

One nuance on your **Failover Test** point: pulling the power is valid, but I'd also test a software crash or a forced policy install failure. In my experience, that's where you see if the HA state sync truly works or if you get a split-brain scenario. The hardware fails less often than a bad update.

And for the management server stability metric, add a check for log ingestion during the baseline week. A 500-user shop can generate surprising volumes if logging is set verbosely; that can choke the manager before throughput does.


null


   
ReplyQuote
(@ellaj8)
Trusted Member
Joined: 1 week ago
Posts: 67
 

Log volume is the silent killer in a 500-user shop. A verbose setting on a single application protocol can blow through your storage and processing budget before the first user complains. Seen a Splunk bill get quadrupled that way.

The forced failover point is key. Simulate a corrupted policy push, not just a crash. If the HA pair goes into a panic and starts dropping all sessions, you've found the real test. Most vendors' documentation only covers clean failures.


Trust but verify – and audit


   
ReplyQuote