Managing Large Networks
In large networks, distributed management systems are required to support scalable
policy administration and log aggregation from a NOC:
Central
Managers:
NetScreen
RapidStream
SonicWALL
Current
Product
NetScreen-Global
PRO 3.0
Central
Policy Management System 3.0
SonicWall
Global Management System 2.1
Versions
and # of Nodes
Express
(25 to 100 nodes) Standard (100 to thousands)
Standard
(10 to 1000 nodes)
Entry
Edition (2 to 20 nodes) or Standard (25 to thousands)
List
Prices
From
$5995 (25 nodes) to $49,995 (1000 nodes), then $11,995 per 1000
From
$4995 (10 nodes) to $24,995 (1000 nodes)
From
$4995 (25 nodes)
Unfortunately, we could not evaluate the high-volume NetScreen and SonicWALL
systems. Instead, we trialed smaller demo versions: NetScreen-Global Manager
and SGMS 1.2. Although our test network was too small to fully-exercise these
complex systems, we can offer a few observations:
Using SonicWALL's GMS, it was easy to push immediate or scheduled changes,
made at a global or group level, to many appliances. An "A-to-B" configurator
made point-to-point VPN tunnel creation simple. Nonetheless, we found ourselves
using the device-level GUI cut-through for many tasks. Recently-integrated
ViewPoint offers extensive log-based reports, but SGMS really focuses on provisioning
rather than fault monitoring.
Using RapidStream's CPM Policy Manager, we defined firewall/VPN/traffic
policies for the entire network, letting CPM derive objects and rules to implement
each device-level policy. Full-mesh and hub-and-spoke VPNs were easy to configure
and reusable objects helped promote consistency. However, derived policies
would be better if they could be fully viewed before deployment. CPM keep-alives
and alarm windows provide network-level surveillance, but CPM lacks persistent
storage for alarms and logs.
Because NetScreen-Global PRO 3.0 is substantially different from the tested
NetScreen-Global Manager, we have no first-hand observations to share. According
to specs, PRO can push centrally-configured policies to many NetScreen appliances,
using role-based privileges to control what each operator can see or do. PRO
can also be integrated with Micromuse Netcool for fault management and root
cause analysis.
Like SGMS and PRO, RapidStream CPM is a client/server system. But the CPM
Server runs on Windows, managing up to 1000 nodes. NetScreen-Global PRO 3.0
and SGMS 2.1 scale to "thousands" of appliances, running on Solaris and Oracle
database servers. PRO Express and SGMS Entry products use smaller platforms,
but top out at just 100 and 25 nodes, respectively. For carriers, the choice
is clear. For smaller regional ISPs, investment in big-league infrastructure
can be harder to justify. Our advice: don't wait until you have 100 appliances
in the field to consider the cost and power of these distributed management
systems.
Network and Firewall Features
These appliances are all ICSA-certified stateful packet inspection firewalls.
Here is a thumbnail comparison of the network and firewall features we found:
Import
Config,
Network & Service Rules, Named Services, Built-In Services
Server
Load Balancing
NetScreen-100
Yes
No
Traffic
Shaping
Bandwidth
per IF
Per Policy: Priority, ToS Bits, Guaranteed, Max Bandwidth
Bandwidth
per IF
Per Policy: Priority, ToS Bits
No
VLAN
Tags
No
Yes
No
Anti-Hacking
Features
Enable
Ports,
Ping per IF
Detect scans,
address sweeps,
Ping of Death
WinNuke, Land, and Tear Drop attacks, as well as ICMP, UDP, and
TCP SYN floods,
malicious URLs
Enable
Ports,
Ping per IF
Detect Ping of Death and IP Source Route attacks, as well as ICMP,
UDP, and TCP SYN packet floods, threshold connect requests to one
server or from one client
Stealth
Mode on WAN IF
Detect scans, Ping of Death, Land, IP Spoof attacks, TCP SYN floods
Routing
Protocols
None
Listen
to RIP, OSPF
None
Other
LAN Services
URL
Filtering Plug-In
DHCP Server
DHCP Relay
DNS Cache
None
URL
Filtering,
AV Plug-Ins,
DHCP Server,
Forward to Web Cache, NetBIOS PassThru
At the low-end of IPRVnet's market, firewalls must be dropped into LANs with
minimal fuss. To facilitate this, these appliances include DHCP/PPPoE assigned
WAN IPs and outgoing NAT/PAT. Other features that are attractive for small accounts:
transparent bridging mode (NetScreen), serving DHCP to LAN hosts (SonicWALL,
NetScreen), plug-in services like URL filtering, and VPN pass-throughs (RapidStream,
SonicWALL).
Slightly larger SMB accounts will look for features like DMZ servers and inbound
address translation (virtual IPs and/or 1-to-1 NAT). Server load balancing is
available on the NetScreen-100, but RapidStream includes load balancing and
VLAN tagging with all its appliances. Only RapidStream lets NAT be applied to
DMZ policies. Only NetScreen supports policy-based NATa complex but valuable
solution that enables tunneling between overlapping private subnets.
As Internet or site-to-site VPN use grows, SMBs often encounter stiff competition
for WAN bandwidth. Traffic management can help customers understand and control
their WAN use. NetScreen and RapidStream appliances can assign bandwidth to
an interface and prioritize individual policies. NetScreen goes further by guaranteeing
and capping bandwidth per policy, graphing results. These integrated RapidStream
and NetScreen features cannot be separately activated or licensed. But they
can help an ISP land or keep an account that's having trouble managing network
utilization.
Policies that permit, deny, block, or authenticate sessions are the cornerstone
of every stateful inspection firewall. Each vendor has a different policy GUI,
making direct comparison difficult. Personal preference and level of expertise
determines which is easiest to use, so there is little point naming our favorite.
Instead, we've listed the policy definition aids that made our life easier during
this eval. For example, RapidStream's built-in address objects ensured that
affected policies were updated whenever an interface was renumbered. NetScreen's
policies are uni-directional, but a matching reverse policy can be created or
updated automatically. SonicWALL's service policies simply have fewer parameters.