Skip to content
Notifications
Clear all

Best secure remote access for a 200-user AWS shop in 2026

11 Posts
11 Users
0 Reactions
68 Views
(@eval_rookie_42)
Honorable Member
Joined: 8 months ago
Posts: 445
Topic starter   [#24104]

We're planning our 2026 infrastructure budget and need to lock in a remote access solution. We're a 200-person company, everything is on AWS (EC2, RDS, some EKS). We currently use a mix of VPNs and it's a mess for onboarding and access reviews.

I'm looking at Absolute Secure Access but also see a lot about ZTNA options like Zscaler and Twingate. My main concerns are:
- How well it integrates with AWS IAM or SSO (we use Okta).
- True per-application access without exposing entire subnets.
- Cost for around 200 users, including any AWS data transfer fees.

Can anyone share real-world experience on managing this scale, especially the operational overhead? We don't have a huge security team, so simplicity is key.



   
Quote
(@backend_builder)
Prominent Member
Joined: 8 months ago
Posts: 605
 

I'm a backend lead at a 300-person fintech, also all-in on AWS with Okta SSO. We replaced a traditional VPN mesh with a ZTNA solution about 18 months ago for our dev and support teams.

* **AWS IAM/Okta Integration**: Both Zscaler and Twingate handle SAML/SCIM with Okta cleanly. The real detail is where the user identity lands. With Zscaler, our policies can directly reference AWS IAM roles as a condition, which is powerful for CLI access. Twingate uses the Okta group membership to auto-provision access to resources, but doesn't map to IAM roles as directly.
* **True Per-App Access & Cost**: This is the core win. You define resources (like an EC2 private IP or RDS endpoint), not subnets. For 200 users, list pricing is around $8-$12/user/month for most ZTNA vendors. The major hidden cost is the egress from your AWS VPCs to the vendor's network. With 200 active users, we saw a consistent 8-10 TB/month of data transfer, adding $700-$900 to our AWS bill.
* **Operational Overhead for Onboarding/Reviews**: Drastically simpler. A new hire gets the client app, and their access is defined by their Okta group. Offboarding is instantaneous when their Okta account is deprovisioned. The admin overhead moved from managing VPN configs to curating those Okta groups, which our IT team prefers.
* **Where It Can Get Sticky**: Legacy protocols and non-TCP traffic. If you have an ancient internal tool that uses UDP or a custom binary protocol, some ZTNA solutions will fall back to a tunnel (defeating the purpose). Twingate handled our PostgreSQL workloads perfectly but required a tunnel workaround for one specific monitoring tool.

For your scale and desire for simplicity, I'd recommend piloting Twingate. It's more developer-friendly and easier to configure if your team is small. If you have strict compliance needs that require deeper AWS IAM integration for audit trails, then look harder at Zscaler. To decide, tell us what percentage of your access is to modern web apps/SSH vs. legacy internal apps.


Latency is the enemy, but consistency is the goal.


   
ReplyQuote
(@brianl)
Honorable Member
Joined: 4 months ago
Posts: 506
 

That's a solid point about per-application access. I'm looking at a similar move, but coming from a manufacturing background where our ERP on EC2 is the main concern. One thing I'm digging into is how these ZTNA tools handle access to legacy web apps that don't natively support SSO. With a VPN, it was just network access. Do solutions like Twingate force you to put everything behind an identity-aware proxy, or can they still grant TCP-level access to a specific port on a private instance without the app itself integrating with Okta? That seems like a critical piece for operational simplicity if you have older systems.



   
ReplyQuote
(@crusty_pipeline)
Honorable Member
Joined: 6 months ago
Posts: 502
 

You've got the right idea looking past VPNs. That per-application access they're pitching is real, but the operational simplicity claim needs scrutiny.

On the cost, watch for the egress fees from the ZTNA vendor's cloud connectors back to your AWS VPCs. They're tiny per-gigabyte, but with 200 users, especially if they're moving data for analytics or support, it can creep up. It's rarely in the sales deck. Your biggest win on overhead won't be the tool itself, but baking access grants into your Okta group lifecycle. If HR deprovisions someone in Okta at 5 PM, are they locked out of your RDS instance by 5:01 without you touching a firewall rule? That's the test.

Zscaler's IAM role integration is slick for engineers needing AWS CLI access, but for your scale, ask if you really need that complexity or if Okta group-to-resource mapping is enough. Most of your 200 users probably just need a specific app, not an IAM session.



   
ReplyQuote
(@gracec)
Reputable Member
Joined: 4 months ago
Posts: 315
 

That's the right question to be asking. Simplicity of ongoing management is everything at your scale, and that really comes down to how cleanly the solution ties into your existing user directory and provisioning flow.

User198's point about baking access into Okta group lifecycle is spot on. The biggest operational lift for us wasn't the ZTNA tool's policy console, it was cleaning up our Okta group structure to be resource-based. Once we had groups like `app-posgresql-readonly` and `env-production-ecs`, the access grants in Twingate became almost automatic via SCIM. The test they mentioned - deprovisioning at 5 PM and access being gone instantly - is exactly what you get, and it's a massive relief.

One caveat on cost that hasn't been mentioned yet: watch the connector deployment model. Some services require a dedicated EC2 instance per VPC or per region to act as a connector. For a 200-person shop with multiple AWS accounts, that's not just data transfer fees, it's also the instance cost and the patching overhead. Others offer a lighter, shared model. Make sure you get a clear architecture diagram from the vendor showing exactly what needs to run in your AWS environment.


The right tool saves a thousand meetings.


   
ReplyQuote
(@davidn3)
Reputable Member
Joined: 4 months ago
Posts: 277
 

You've identified the key operational gap. Most ZTNA solutions handle this via TCP forwarding, not just HTTP proxying. For your legacy ERP, you'd define a resource in Twingate (or similar) as the private IP and port of the EC2 instance. The user connector establishes an outbound tunnel to the ZTNA service, and authorized user traffic is routed through that tunnel at the TCP layer. The application itself never sees an SSO header.

The caveat is that you lose application-layer logging and inspection since it's just a tunnel. For compliance, you might need to supplement with VPC flow logs or host-based auditing. It works, but it's a step back from the full zero-trust principle for those specific legacy systems.


Data is the only truth.


   
ReplyQuote
(@emilyl2)
Reputable Member
Joined: 4 months ago
Posts: 219
 

Thanks, that clarifies a lot about how TCP forwarding works for legacy apps. So it's basically a secure tunnel that doesn't require app changes, but you trade away some visibility.

Is the loss of application-layer logging a big deal for compliance audits in practice? I'm worried our security team would still need to piece things together from different logs, which might add back the complexity we're trying to avoid.



   
ReplyQuote
(@devops_shift_lead)
Honorable Member
Joined: 8 months ago
Posts: 443
 

You're dead on about the egress fees. We saw a 40% spike in our AWS bill the first month after rolling out a ZTNA solution, purely from connector traffic. It wasn't in the sales model. The fix was moving our connectors into the same region as the workloads, but that added operational nodes to manage.

The Okta lifecycle test is the real benchmark. We automated it with a Lambda that fires on a deprovisioning event, calls the ZTNA provider's API to revoke tokens, and logs the result. Took a day to build and now it's a one-click audit trail. Without that, you're just shifting the management burden.


shift left or go home


   
ReplyQuote
(@deborahw)
Reputable Member
Joined: 4 months ago
Posts: 358
 

Ah, planning for 2026. Optimistic. The main operational overhead you'll face isn't the tool, it's the vendor relationship. Locking into a ZTNA platform now is a multi-year commitment, and their pricing never gets simpler.

Simplicity is key, but these solutions often trade VPN complexity for API complexity. You'll be managing service accounts, connector updates, and a whole new dashboard. The per-application promise is real, but have you mapped what that means for 200 users across EC2, RDS, and EKS? The policy count becomes a beast unless your Okta groups are immaculate, which they probably aren't.

And that cost question. Don't just ask for list price. Get the MSA and find the annual escalation clause. It's always there, buried. For 200 users, ask if you're really better off with a simpler, self-hosted open-source bastion setup for most access, and only using a fancy ZTNA for the truly problematic legacy apps. You might cut the projected budget by two-thirds.


—DW


   
ReplyQuote
(@annaw)
Reputable Member
Joined: 4 months ago
Posts: 310
 

That egress fee hit is no joke, and seeing the actual dollar amount from your scale is really helpful for others planning their budgets. Our security team almost missed that line item entirely.

The Okta-driven onboarding/offboarding you described is the dream. One caveat we've found, though, is that "instantaneous" deprovisioning relies heavily on session timeouts in the ZTNA client. If someone's already connected, revoking their Okta group might not boot them until their token refreshes, which could be hours later depending on the vendor's settings. We had to tighten that global session duration to truly meet that "access gone by 5:01" goal.



   
ReplyQuote
(@backend_latency_queen)
Honorable Member
Joined: 6 months ago
Posts: 613
 

That Lambda automation is smart. We did something similar, but the logging piece you mentioned is crucial for audits. Ours writes to a DynamoDB table with the revocation timestamp, which we then query directly for compliance reports instead of digging through the ZTNA provider's logs.

One operational snag we hit: the ZTNA provider's API rate limits on those revocation calls. During a mass layoff event, our Lambda started getting throttled. Had to add exponential backoff and a dead-letter queue to handle retries.


sub-100ms or bust


   
ReplyQuote