WAN optimization concepts

Client/server architecture

Traffic across a WAN typically consists of clients on a client network communicating across a WAN with a remote server network. The clients do this by starting communication sessions from the client network to the server network. These communication sessions can be open text over the WAN or they can be encrypted by SSL VPN or IPsec VPN.

To optimize these sessions, you can add WAN optimization security policies to the client-side FortiGate unit to accept sessions from the client network that are destined for the server network. The client-side FortiGate unit is located between the client network and the WAN. WAN optimization security policies include WAN optimization profiles that control how the traffic is optimized.

The client-side FortiGate unit must also include the IP address of the server-side FortiGate unit in its WAN optimization peer configuration. The server-side FortiGate unit is located between the server network and the WAN, The peer configuration allows the client-side FortiGate unit to find the server-side FortiGate unit and attempt to establish a WAN optimization tunnel with it.

For the server-side FortiGate unit you must add a security policy with wanopt as the Incoming Interface. This security policy allows the FortiGate unit to accept WAN optimization sessions from the client-side FortiGate unit. For the server-side FortiGate unit to accept a WAN optimization connection it must have the client-side FortiGate unit in its WAN optimization peer configuration.

WAN optimization profiles are only added to the client-side WAN optimization security policy. The server-side FortiGate unit employs the WAN optimization settings set in the WAN optimization profile on the client-side FortiGate unit.

Client/server architecture

When both peers are identified the FortiGate units attempt to establish a WAN optimization tunnel between them. WAN optimization tunnels use port 7810. All optimized data flowing across the WAN between the clientside and server-side FortiGate units use this tunnel. WAN optimization tunnels can be encrypted use SSL encryption to keep the data in the tunnel secure.

Any traffic can be sent through a WAN optimization tunnel. This includes SSL and IPsec VPN traffic. However, instead of configuring SSL or IPsec VPN for this communication you can add SSL encryption using the WAN optimization tunnel.

In addition to basic identification by peer host ID and IP address you can configure WAN optimization authentication using certificates and pre-shared keys to improve security. You can also configure FortiGate units involved in WAN optimization to accept connections from any identified peer or restrict connections to specific peers.

The FortiClient application can act in the same manner as a client-side FortiGate unit to optimize traffic between a computer running FortiClient and a FortiGate unit.

WAN optimization peers

The client-side and server-side FortiGate units are called WAN optimization peers because all of the FortiGate units in a WAN optimization network have the same peer relationship with each other. The client and server roles just relate to how a session is started. Any FortiGate unit configured for WAN optimization can be a client-side and a server-side FortiGate unit at the same time, depending on the direction of the traffic. Client-side FortiGate units initiate WAN optimization sessions and server-side FortiGate units respond to the session requests. Any FortiGate unit can simultaneously be a client-side FortiGate unit for some sessions and a server-side FortiGate unit for others.

WAN optimization peer and tunnel architecture

To identify all of the WAN optimization peers that a FortiGate unit can perform WAN optimization with, you add host IDs and IP addresses of all of the peers to the FortiGate unit configuration. The peer IP address is actually the IP address of the peer unit interface that communicates with the FortiGate unit.

Protocol optimization

Protocol optimization techniques optimize bandwidth use across the WAN. These techniques can improve the efficiency of communication across the WAN optimization tunnel by reducing the amount of traffic required by Protocol optimization and MAPI

communication protocols. You can apply protocol optimization to Common Internet File System (CIFS), FTP, HTTP, MAPI, and general TCP sessions. You can apply general TCP optimization to MAPI sessions.

For example, CIFS provides file access, record locking, read/write privileges, change notification, server name resolution, request batching, and server authentication. CIFS is a fairly “chatty” protocol, requiring many background transactions to successfully transfer a single file. This is usually not a problem across a LAN. However, across a WAN, latency and bandwidth reduction can slow down CIFS performance.

When you select the CIFS protocol in a WAN optimization profile, the FortiGate units at both ends of the WAN optimization tunnel use a number of techniques to reduce the number of background transactions that occur over the WAN for CIFS traffic.

If a policy accepts a range of different types of traffic, you can set Protocol to TCP to apply general optimization techniques to TCP traffic. However, applying this TCP optimization is not as effective as applying more protocolspecific optimization to specific types of traffic. TCP protocol optimization uses techniques such as TCP SACK support, TCP window scaling and window size adjustment, and TCP connection pooling to remove TCP bottlenecks.

Protocol optimization and MAPI

By default the MAPI service uses port number 135 for RPC port mapping and may use random ports for MAPI messages. The random ports are negotiated through sessions using port 135. The FortiOS DCE-RPC session helper learns these ports and opens pinholes for the messages. WAN optimization is also aware of these ports and attempts to apply protocol optimization to MAPI messages that use them. However, to configure protocol optimization for MAPI you should set the WAN optimization profile to a single port number (usually port 135). Specifying a range of ports may reduce performance.

Byte caching

Byte caching breaks large units of application data (for example, a file being downloaded from a web page) into small chunks of data, labeling each chunk of data with a hash of the chunk and storing those chunks and their hashes in a database. The database is stored on a WAN optimization storage device. Then, instead of sending the actual data over the WAN tunnel, the FortiGate unit sends the hashes. The FortiGate unit at the other end of the tunnel receives the hashes and compares them with the hashes in its local byte caching database. If any hashes match, that data does not have to be transmitted over the WAN optimization tunnel. The data for any hashes that does not match is transferred over the tunnel and added to that byte caching database. Then the unit of application data (the file being downloaded) is reassembled and sent to its destination.

The stored byte caches are not application specific. Byte caches from a file in an email can be used to optimize downloading that same file or a similar file from a web page.

The result is less data transmitted over the WAN. Initially, byte caching may reduce performance until a large enough byte caching database is built up.

To enable byte caching, you select Byte Caching in a WAN optimization profile.

Byte caching cannot determine whether or not a file is compressed (for example a zip file), and caches compressed and non-compressed versions of the same file separately.

Dynamic data chunking for byte caching

Dynamic data chunking can improve byte caching by improving detection of data chunks that are already cached in changed files or in data embedded in traffic using an unknown protocol. Dynamic data chunking is available for HTTP, CIFS and FTP.

Use the following command to enable dynamic data chunking for HTTP in the default WAN optimization profile.

config wanopt profile edit default config http set prefer-chunking dynamic

end

By default dynamic data chunking is disabled and prefer-chunking is set to fix.

WAN optimization transparent mode

WAN optimization is transparent to users. This means that with WAN optimization in place, clients connect to servers in the same way as they would without WAN optimization. However, servers receiving packets after WAN optimization “see” different source addresses depending on whether or not transparent mode is selected for WAN optimization. If transparent mode is selected, WAN optimization keeps the original source address of the packets, so servers appear to receive traffic directly from clients. Routing on the server network should be configured to route traffic with client source IP addresses from the server-side FortiGate unit to the server and back to the server-side FortiGate unit.

Some protocols, for example CIFS, may not function as expected if transparent mode is not selected. In most cases, for CIFS WAN optimization you should select transparent mode and make sure the server network can route traffic as described to support transparent mode.

If transparent mode is not selected, the source address of the packets received by servers is changed to the address of the server-side FortiGate unit interface that sends the packets to the servers. So servers appear to receive packets from the server-side FortiGate unit. Routing on the server network is simpler in this case because client addresses are not involved. All traffic appears to come from the server-side FortiGate unit and not from individual clients.

Do not confuse WAN optimization transparent mode with FortiGate transparent mode. WAN optimization transparent mode is similar to source NAT. FortiGate’s transparent mode is a system setting that controls how the FortiGate unit (or a VDOM) processes traffic.

Configuring transparent mode

You can configure transparent mode by selecting Transparent in a WAN Optimization profile. The profile is added to an active WAN Optimization policy.

When you configure a passive WAN Optimization policy you can accept the active policy transparent setting or you can override the active policy transparent setting. From the GUI you can do this by setting the Passive Option as follows:

• default use the transparent setting in the WAN Optimization profile added to the active policy (client-side configuration).
• transparent impose transparent mode (override the active policy transparent mode setting). Packets exiting the FortiGate keep their original source addresses.
• non-transparent impose non-transparent mode (override the active policy transparent mode setting). Packets exiting the FortiGate have their source address changed to the address of the server-side FortiGate unit interface that sends the packets to the servers.

From the CLI you can use the following command:

config firewall policy set wanopt-passive-opt {default | transparent | non-transparent}

end

Operating modes and VDOMs

To use WAN optimization, the FortiGate units can operate in either NAT/Route or transparent mode. The clientside and server-side FortiGate units do not have to be operating in the same mode.

As well, the FortiGate units can be configured for multiple virtual domain (VDOM) operation. You configure WAN optimization for each VDOM and configure one or both of the units to operate with multiple VDOMs enabled.

If a FortiGate unit or VDOM is operating in transparent mode with WAN optimization enabled, WAN optimization uses the management IP address as the peer IP address of the FortiGate unit instead of the address of an interface.

WAN optimization tunnels

All optimized traffic passes between the FortiGate units over a WAN optimization tunnel. Traffic in the tunnel can be sent in plain text or encrypted using AES-128bit-CBC SSL.

WAN optimization tunnels

Both plain text and the encrypted tunnels use TCP destination port 7810.

Before a tunnel can be started, the peers must be configured to authenticate with each other. Then, the clientside peer attempts to start a WAN optimization tunnel with the server-side peer. Once the peers authenticate with each other, they bring up the tunnel and WAN optimization communication over the tunnel starts. After a tunnel has been established, multiple WAN optimization sessions can start and stop between peers without restarting the tunnel.

Tunnel sharing

You can use the tunnel-sharing WAN optimization profile CLI keyword to configure tunnel sharing for WAN optimization rules. Tunnel sharing means multiple WAN optimization sessions share the same tunnel. Tunnel sharing can improve performance by reducing the number of WAN optimization tunnels between FortiGate units. Having fewer tunnels means less data to manage. Also, tunnel setup requires more than one exchange of information between the ends of the tunnel. Once the tunnel is set up, each new session that shares the tunnel avoids tunnel setup delays.

Tunnel sharing also uses bandwidth more efficiently by reducing the chances that small packets will be sent down the tunnel. Processing small packets reduces network throughput, so reducing the number of small packets improves performance. A shared tunnel can combine all the data from the sessions being processed by the tunnel and send the data together. For example, suppose a FortiGate unit is processing five WAN optimization sessions and each session has 100 bytes to send. If these sessions use a shared tunnel, WAN optimization combines the packets from all five sessions into one 500-byte packet. If each session uses its own private tunnel, five 100-byte packets will be sent instead. Each packet also requires a TCP ACK reply. The combined packet in the shared tunnel requires one TCP ACK packet. The separate packets in the private tunnels require five.

Use the following command to configure tunnel sharing for HTTP traffic in a WAN optimization profile.

config wanopt profile edit default config http set tunnel-sharing {express-shared | private | shared}

end

Tunnel sharing is not always recommended and may not always be the best practice. Aggressive and nonaggressive protocols should not share the same tunnel. An aggressive protocol can be defined as a protocol that is able to get more bandwidth than a non-aggressive protocol. (The aggressive protocols can “starve” the non-

WAN optimization and user and device identity policies, load balancing and traffic shaping

aggressive protocols.) HTTP and FTP are considered aggressive protocols. If aggressive and non-aggressive protocols share the same tunnel, the aggressive protocols may take all of the available bandwidth. As a result, the performance of less aggressive protocols could be reduced. To avoid this problem, rules for HTTP and FTP traffic should have their own tunnel. To do this, set tunnel-sharing to private for WAN optimization rules that accept HTTP or FTP traffic.

It is also useful to set tunnel-sharing to express-shared for applications, such as Telnet, that are very interactive but not aggressive. Express sharing optimizes tunnel sharing for Telnet and other interactive applications where latency or delays would seriously affect the user’s experience with the protocol.

Set tunnel-sharing to shared for applications that are not aggressive and are not sensitive to latency or delays. WAN optimization rules set to sharing and express-shared can share the same tunnel.

WAN optimization and user and device identity policies, load balancing and traffic shaping

Please note the following about WAN optimization and firewall policies:

• WAN optimization is not compatible with firewall load balancing.
• WAN optimization is compatible with source and destination NAT options in firewall policies (including firewall virtual IPs). If a virtual IP is added to a policy the traffic that exits the WAN optimization tunnel has its destination address changed to the virtual IPs mapped to IP address and port.
• WAN optimization is compatible with user identity-based and device identity security policies. If a session is allowed after authentication or device identification the session can be optimized.

Traffic shaping

Traffic shaping works for WAN optimization traffic that is not in a WAN optimization tunnel. So traffic accepted by a WAN optimization security policy on a client-side FortiGate unit can be shaped on ingress. However, when the traffic enters the WAN optimization tunnel, traffic shaping is not applied.

In manual mode:

• Traffic shaping works as expected on the client-side FortiGate unit. l Traffic shaping cannot be applied to traffic on the server-side FortiGate unit.

In active-passive mode:

• Traffic shaping works as expected on the client-side FortiGate unit.
• If transparent mode is enabled in the WAN optimization profile, traffic shaping also works as expected on the server-side FortiGate unit. l If transparent mode is not enabled, traffic shaping works partially on the server-side FortiGate unit.

WAN optimization and HA

You can configure WAN optimization on a FortiGate HA cluster. The recommended best practice HA configuration for WAN optimization is active-passive mode. When the cluster is operating, all WAN optimization WAN optimization, web caching and memory usage

sessions are processed by the primary unit only. Even if the cluster is operating in active-active mode, HA does not load-balance WAN optimization sessions.

You can also form a WAN optimization tunnel between a cluster and a standalone FortiGate unit or between two clusters.

In a cluster, only the primary unit stores the byte cache database. This database is not synchronized to the subordinate units. So, after a failover, the new primary unit must rebuild its byte cache. Rebuilding the byte cache can happen relatively quickly because the new primary unit gets byte cache data from the other FortiGate unit that it is participating with in WAN optimization tunnels.

WAN optimization, web caching and memory usage

To accelerate and optimize disk access and to provide better throughput and less latency FortiOS WAN optimization uses provisioned memory to reduce disk I/O and increase disk I/O efficiency. In addition, WAN optimization requires a small amount of additional memory per session for comprehensive flow control logic and efficient traffic forwarding.

When WAN optimization is enabled you will see a reduction in available memory. The reduction increases when more WAN optimization sessions are being processed. If you are thinking of enabling WAN optimization on an operating FortiGate unit, make sure its memory usage is not maxed out during high traffic periods.

In addition to using the system dashboard to see the current memory usage you can use the get test wad 2 command to see how much memory is currently being used by WAN optimization. See “get test {wad | wccpd} <test_level>” for more information.