Now that we’ve been through technical architecture considerations for the evolving firewall (Part 1, Part 2), let’s talk about deployment considerations. Depending on requirements, there many different ways to deploy enterprise firewalls. Do this wrong and you end up with either too many or too few boxes, single points of failure, suboptimal network access, and/or crappy application performance.

We could talk about all sorts of different models and use fancy names like tiered, mesh, peer to peer, and the like for them – but fortunately the situation isn’t really that complicated. To choose the most appropriate architecture you must answer a few questions:

  • Public or Private Network? Are your remote locations all connected via private connections such as MPLS or managed IP services, or via public Internet services leveraging site-to-site VPN tunnels?
  • How much is avoiding downtime worth? This fairly simple question will drive both network architecture and perimeter device selection. You can implement high availability architectures to minimize the likelihood of downtime but the cost is generally significant.
  • What egress filtering/protection do you need? Obviously you want to provide web and email filtering on outbound traffic. Depending on bandwidth availability and cost, it may make sense to haul that back to a central location to be processed by large (existing) content security gateways. But for bandwidth-constrained sites it may make more sense to do web/email filtering locally (using a UTM box), with the understanding that filtering at the smaller sites might be less sophisticated.
  • Who controls gateway policy? Depending on the size of your organization, there may be different policies for different geographies, business units, locations, etc. Some enterprise firewall management consoles support this kind of granular policy distribution, but you need to figure out who will control policy, and use this to guide how you deploy the boxes.

Remember the technical architecture post where we pointed out the importance of consistency. A consistent feature set on devices up and down a vendor’s product line provides a lot of flexibility in how you can deploy – this enables you to select equipment based on the throughput requirement rather than feature set. This is also preferable because application architectures and requirements change, and support for all features on branch equipment (even if you don’t initially expect to use them) saves deploying new equipment remotely if you decide to take advantage of those features later, but we recognize this is not always possible. Economic reality rears its head every so often.

Bandwidth Matters

We most frequently see tiers of firewalls implemented in either two or three tiers. Central sites (geographic HQ) get big honking firewalls deployed in a high-availability cluster configuration to ensure resilience and throughput – especially if they provide higher-level application and/or UTM features. Distribution locations, if they exist, are typically connected to the central site via a private IP network. These tend to be major cities with good bandwidth. With plentiful bandwidth, most organizations tend to centralize egress filtering to minimize the control points, so outbound traffic tends to be centralized through the central site.

With smaller locations like stores, or in emerging countries with expensive private network options, it may make more economic sense to use public IP services (commodity Internet access) with site-to-site VPN. In this case it’s likely not performance (or cost) effective to centralize egress filtering, so these firewalls generally must do the filtering as well.

Regardless of the egress filtering strategy you should have a consistent set of ingress policies in place, which usually means (almost) no traffic originating from the Internet is accepted: a default deny security posture. Most organizations leverage hosting providers for web apps, which allow tight rules to be placed on the perimeter for inbound traffic. Likewise, allowing inbound Internet traffic to a small location usually doesn’t make sense, since those small sites shouldn’t be directly serving up data. Unless you are cool with tellers running their Internet-based side businesses on your network.

High Availability Clusters

Downtime is generally a bad thing – end users can get very grumpy when they can’t manage their fantasy football teams during the work day – so you should investigate the hardware resilience features of firewall devices. Things like hot swappable drives and power supplies, redundant backplanes, multiple network connections, redundant memory, etc. Obviously the more redundancy built into the box, the more it will cost, but you already knew that.

Another option is to deploy a high availability cluster. Basically, this means you’ve got two (or more) boxes using sharing a single configuration, allowing automated and transparent load balancing between them to provide stable the performance and ride out any equipment failures. So if a box fails its peer(s) transparently pick up the slack.

High availability and clustering used to be different capabilities (and on some older firewall architectures, still are). But given the state of the hardware and maturity of the space, the terminology has evolved to active/active (all boxes in the cluster process traffic) and active/passive (some boxes are normally hot spares, not processing traffic). Bandwidth requirements tend to drive whether multiple gateways are active, but the user-visible functioning is the same.

Internal Deployment

We have mostly discussed the perimeter gateway use case. But there is another scenario, where the firewall is deployed within the data center or at distribution points in the network, and provides network segmentation and filtering. This is a bit different than managing inbound/outbound traffic at the perimeter, and largely driven by network architecture. The bandwidth requirements for internal devices are intense – typically 40-100gbps and here downtime is definitely a no-no, so provision these devices accordingly and bring your checkbook.


The final issue we’ll tackle in relation to deployment is getting old boxes out and new boxes in. Depending on the size of the environment, it may not be feasible to do a flash cutover. So the more the new vendor can do to assist in the migration, the better. Fortunately the market is mature enough that many vendors can read in their competitors’ rule sets, which can be facilitate switchovers.

But don’t forget that a firewall migration is normally a great opportunity to revisit the firewall rule base and clear out the crap. Yes, as we discussed in the Network Security Ops Quant research, you want to revisit your policies/rules systematically (hopefully a couple times a year), but we are realists. Having to update rules for new capabilities within new gear provides both the means and the motive to kill some of those stale firewall rules.

We’re about halfway through the Selection process. Next we’ll tackle enterprise firewall management expectations before moving on to the advanced features that really differentiate these devices.