Virtual Server & Server Load Balancing

Virtual Server is a method for single gateway machine to act as multiple servers while the real servers sit inside corporate network to process requests passed in from the gateway machine. Inbound traffic does not have to know where the real servers are, or whether there are just one or many servers. This method prevents direct access by users and therefore increases security and flexibility.

FortiWAN has built in virtual server and is capable of supporting various virtual server mapping methods. For example, different public IP addresses can be mapped to various real servers in LAN or DMZ. Or ports can be mapped to public IP address on different servers.

Virtual server are configured by designating and adjusting virtual server rules. Each rule specifies a mapping condition. It maps WAN IP address and a service (port or ports) to an internal server IP. The order of virtual server rules is like any other rule tables in FortiWAN as it also uses the “first match scheme”, viz. the first rule of request matched is the rule to take effect.

For example, a public IP address 211.21.48.196 and wants a web server on 192.168.123.16 to handle all the web page requests coming to this public IP address. To do this, a virtual server rule must be created with 211.21.48.196 to be its WAN IP, 192.168.123.16 to be its Server IP, and HTTP(80) to be its Service.

Virtual Server makes intranet (LAN) servers accessible for the internet (WAN). The private IP addresses assigned to intranet servers will become invisible to the external environment, making services accessible for users outside the network. Then FortiWAN is available to redirect these external requests to the servers in LAN or DMZ. Whenever an external request arrives, FortiWAN will consult the Virtual Server table and redirect the packet to the corresponding server in LAN or DMZ. The rules of Virtual Server tables are prioritized top down. If one rule is similar to another in the table, only the higher ranked one will be applied, and the rest will be ignored. In addition, Virtual Server enables to balance load on multiple servers, which is to distribute traffic over a group of servers (server cluster), making services highly accessible.

Virtual Server & Server Load Balancing

FortiWAN provides mechanisms to record, notify and analysis on events refer to the Virtual Server service, see “Log”, “Statistics: Virtual Server Status” and “Report: Virtual Server”.

IPv4 Virtual Server

IPv6 Virtual Server

Example 1

The settings for virtual servers look like:

• Assign IP address 211.21.48.194 to WAN1. Refer to [System] -> [Network Settings] -> [WAN Settings] for more regarding WAN IP configurations. l Assign IP address 211.21.33.186 to WAN2.

Virtual Server & Server Load Balancing

• Forward all HTTP requests (port 80) through WAN1 or WAN2 to the two HTTP servers 192.168.0.100 and 192.168.0.101 in LAN.
• Forward all FTP requests (port 21) through WAN1 or WAN2 to two FTP servers 192.168.0.200 and 192.168.0.201 in LAN.
• Assign 211.21.48.195 and 211.21.33.189 to WAN 1 and WAN2. Forward all requests to 211.21.48.195 or 211.21.33.189 to two SMTP servers 192.168.0.200 and 192.168.0.201 in LAN. l Forward all requests from 211.21.48.197 to 192.168.0.15 in LAN.

Note:

1. FortiWAN can auto-detect both active and passive FTP servers.
2. All public IPs must be assigned to WAN 1. To configure these IPs, go to “IP(s) on Localhost of the Basic Subnet” table in [System] -> [Network Settings] -> [WAN Settings] -> [WAN Link 1].
3. 21.48.197 does not belong to any physical host, and it must be assigned to WAN port.

Virtual server table for the above settings:

Example 2

The settings for virtual servers look like:

• Forward all the TCP port 1999 requests established between external network and public IP 211.21.48.194 to FTP Server@ TCP port 1999 at 192.168.0.100 in LAN.
• Note: Due to the nature of ftp protocol, in port style ftp-data connection, when ftp-control is used in port 1999, port 1998 will be taken by ftp-data.
• Enable external users to access WAN IP 211.21.33.186, and connect PcAnywhere to .LAN hosts. l Note: PcAnywhere uses TCP port 5631 and UDP port 5632. Refer to PcAnywhere software manual for more details.
• Enable external users to access WAN IP 211.21.48.194, and forward packets of TCP/UDP range 2000-3000 to host 192.168.0.15.

Note: Port range redirecting is supported as well.

Virtual server table for the settings above:

WAN Link Health Detection

Tunnel Routing – Benchmark

To guarantee a performance aggregation transferring TR packets, FortiWAN requires equal quality for the WAN links employed in a tunnel group. The Benchmark here provides evaluation of WAN link quality for every single tunnel. Tunnels are judged in run trip time, packet loss and bandwidth. It is not suggested to employ a WAN link that is worse than others in a tunnel group.

Tunnel Routing’s Benchmark works as Client/Server mode. Test traffic is sent from the client site to the server site via every single configured tunnel, and then the benchmark results are reported at client site. Two steps to start Tunnel Routing’s Benchmark between two FortiWAN appliances (make sure the Tunnel Routing network is established between the two FortiWANs),

1. Specify one of the FortiWANs to be the benchmark server.
2. Start benchmark traffic from the benchmark client, the ForiWAN opposite to the benchmark server.

Start a benchmark server

From the WeB UI, the Tunnel Routing page, all the configured tunnel groups are listed in the Benchmark panel. To start the benchmark server on a FortiWAN for a tunnel group, you need:

1. Specify the port number on the Test Port field for sending/receiving the testing traffic. Note that the port number on both benchmark sites (Client/Server) must be identical. It will fail to receive testing packets if unequal port numbers are used by the two sites.
2. Click the button Start Test Server of the tunnel group that you want to test from the list (in Test Client Status block). This button will be switched to Stop Test Server while benchmark server is running; click it to stop the server.

While the benchmark server is running, a message Test server is running. Please do not change to another page or close browser will display and occupy the main page of Web UI. For all the administrator accounts, it become unable to apply new configurations to Tunnel Routing (the Apply button on Web UI becomes ineffective) during benchmark server is running. Web UI will allow apply configurations to other functions during benchmark server is running, but we suggest not to do this since changes to some functions such as Network Setting, Firewall or IPSec might interrupt benchmark server. During benchmark server running, you can switch Web UI main page to other functions, but a message Test server is running. Please stop it first displays when you turn the main page back to Tunnel Routing. This message reminds you the benchmark server is still running, and the Apply button of Tunnel Routing remains ineffective until you stop the server. Note that the benchmark server can work for only one tunnel group anytime; stop the server on one tunnel group to start it for another.

Start testing traffic from the benchmark client

For the symmetric FortiWAN sites of a tunnel routing network, benchmark client, the site that is opposite to the benchmark server, triggers the testing traffic. Similarly, all the configured tunnel groups are listed in Benchmark panel. To start benchmark traffic on the site you need:

1. Specify the port number on the Test Port field for sending/receiving the test traffic. Note that the port number on both benchmark sites (Client/Server) must be identical. It will fail to receive testing packets if unequal port numbers are used by the two sites.
2. Click the button Test of the same tunnel group that the opposite benchmark server is working for. You will be direct to a management panel to start benchmark testing. For a disable tunnel group, a error message This group is not enabled
3. In the testing management panel, you see all the tunnels of the tunnel group listed (IP addresses of the two endpoints of a tunnel), and two test cases provided:
   1. Single tunnel test: Click the Test button of a tunnel, testing traffic will be generated and sent to the opposite (the server side) of the tunnel. All the packets of the testing session will be sent through only the specified tunnel. This will bring out a testing result for evaluating performance of the specified tunnel.
   2. Tunnel group test: Click the Test button of the last item All Tunnels in Group (at the bottom of the table), testing traffic will be generated and sent to the opposite (the server side) of the tunnel group. All the packets of the testing session will be distributed over the tunnels of the tunnel group according to the configured algorithm of the tunnel group. This will bring out a testing result for evaluating performance of the tunnel group.
4. On the upper right corner of the table, there is a button Test All used to perform every Single Tunnel Testing and the Tunnel Group Testing one by one in a top-down order.
5. You can click Close to stop and leave the benchmark management panel.

How to set up routing rules for Tunnel Routing

To perform Tunnel Routing, symmetric FortiWAN deployment is a basic requirement. Therefore, symmetric routing rules are also required for two-way data transmission. A routing rule here contains three basic elements that are

What is the traffic to be transferred by Tunnel Routing? Tunnel Routing filter traffic by Source, Destination and Service.

Which Tunnel Group is employed to transfer the traffic? Apply a predefined tunnel group to the specified traffic, then it will be transferred according to the how the tunnel group is defined; the balancing algorithm, the tunnels, the weight, the encryption and DSCP.

What to do if the Tunnel Group fails? A failed tunnel group means all the tunnels defined in the tunnel group are disconnected (detected by Tunnel Routing’s tunnel healthy detection mechanism). Therefore, it is necessary to specify another way for the traffic. Note that as long as one tunnel in a tunnel group remains connected, Tunnel Routing keeps employing the tunnel group for transmission.

Next we introduce the two ways, Routing Rule and Default Rule, to establish the routing rules for Tunnel Routing.

Routing Rules

This is the general way to set routing rules for Tunnel Routing. A routing rule contains the three basic elements above, which evaluates traffic by Source, Destination, Service, (Tunnel) Group and Fail-Over. Note that a routing rule sat on a FortiWAN site is required symmetrically for the opposite FortiWAN site, so that the bidirectional transmission is achieved.

If your TR network deployment requires more than 100 TR routing rules, replacing the TR routing rules with TR default rules will be suggested for better performance.

Tunnel Routing – Setting

There are two major steps to set up Tunnel Routing, define the association of tunnels (see the tables: Basic Setting and Tunnel Group) and set up the routing rules (see the tables: Default Rules, Routing Rules and Persistent Rules). Tunnel Routing works in symmetric FortiWAN sites, when the unit we are talking about or configuring to is called local host (or local site), the opposite unit is then called remote host (or remote site).

Basic Setting

The basic settings are located here: enabling or disabling Tunnel Route logging, define names and entering tunnel routing activation key (if the encryption function is enabled for a tunnel group).

Tunnel Routing

Tunneling is a technique to perform data transmission for a foreign protocol over a incompatible network; such as running IPv6 over IPv4, and the transmission of data for use within a private, corporate network through a public network. Tunneling is done by encapsulating and decapsulating data and information of the particular protocol within the incompatible transmission units symmetrically.

Traditional tunneling is established over single WAN link which is a lack of load balancing and fault tolerance. FortiWAN’s Tunnel Routing (TR) is a technique that builds a special connection between two FortiWAN units to deliver link aggregation and fault tolerance over multiple WAN links ideally tailored for multinational intranet systems. Different to Auto Routing distributing sessions over WAN links, Tunnel Routing breaks further a session down to packets over multiple WAN links and allows data to be prioritized during transfer while boosting the performance of critical services such as VPN and live video streaming while avoiding delays and data loss.

Basically, FortiWAN’s Tunnel Routing implies routing packets of a session over tunnels (WAN links), which contains the two elements – Tunnels and Routing.

GRE Tunnel

FortiWAN’s Tunnel Routing sets up proprietary tunnels between symmetric FortiWAN sites (local and remote) with GRE (Generic Routing Encapsulation) protocol. GRE (Generic Routing Encapsulation) Protocol packs the Payload (Original Packet) with Delivery Header and GRE Encapsulation Header. Physically, a point-to-point GRE tunnel for Tunnel Routing is the transimission of GRE packets via a pair of WAN links predefined on the symmetric FortiWAN sites (a WAN link on the local FortiWAN, and another one on the remote FortiWAN) (See “Tunnel Group” and “Group Tunnel” in “Tunnel Routing – Setting”).

Routing

With the multiple WAN links on each FortiWAN, Tunnel Routing distributes (routes) GRE packets of a session over the GRE tunnels (a tunnel group) according the balancing algorithms and tunnel status detection. This is what the load balancing and fault tolerance Tunnel Routing provides for tunneling. Moreover, with proper policy setting, Tunnel Routing can route GRE packets over multiple sites (more than two sites) without full-mesh connections between the sites (See “Default Rule”, “Routing Rule” and “Persistent Rules” in “How to set up routing rules for Tunnel Routing”). Briefly, it performs routing of GRE packets over multiple tunnels and multiple sites.

Next we introduce Tunnel Routing in the following topics:

How the Tunnel Routing Works

Tunnel Routing – Setting

How to set up routing rules for Tunnel Routing

Tunnel Routing – Benchmark

Scenarios

How the Tunnel Routing Works

Here is an example to explain the processes that how Tunnel Routing delivers packets to remote private internal network via Internet. Here are two FortiWAN sites (FWN-A and FWN-B) connected to Internet with two WAN links respectively. Two private LAN networks: 192.168.10.0/255.255.255.0 and 192.168.20.0/255.255.255.0 are connected to FWN-A and FWN-B respectively. Now host 192.168.10.100 would like to communicate with host 192.168.20.100 which is located at remote private LAN. Here are the steps:

1. Host 19.168.10.100 sends the first original packet to FWN-A, source IP and destination IP of the packet are indicated as 192.168.10.100 and 192.168.20.100.
2. FWN-A’s Tunnel Routing takes charge of transferring the packet because it matches a tunnel routing rule (A routing rule is predefined for packets from 192.168.10.0/255.255.255.0 to 192.168.20.0/255.255.255.0).
3. According the specified balancing algorithm (determining a WAN link for transferring), FWN-A encapsulates the original packet with GRE and Delivery headers which the source IP and destination IP are indicated as public addresses 1.1.1.1 (FWN-A’s WAN 1) and 3.3.3.3 (FWN-B’s WAN 1) respectively.
4. The GRE packet is then transferred via Tunnel 1 (from FWN-A’s WAN 1 to FWN-B’s WAN 1 via Internet).
5. FWN-B receives this GRE packet and decapsulates it to recover the original packet.
6. The original packet then is forwarded to host 192.168.20.100 in the private LAN network.
7. The subsequent packets (for example the packet 2 in the figure below) of the session from host 192.168.10.100 are transferred in the same way except the different tunnels that balancing algorithm determines.

After the basic concept how Tunnel Routing transfers packets, several topics related to Tunnel Routing are explained in detail.

Priority over Auto Routing and NAT

Tunnel Routing rules are in higher priority than Auto Routing rules and NAT rules for FortiWAN matching packets with. Predefine a Tunnel Routing rule, a Auto Routing rule (See “Auto Routing”) and a NAT rule (See “NAT”) with the same source and destination, packets that are indicated the source and destination will be first matched to the Tunnel Routing rule and transferred by Tunnel Routing, without be processed by FortiWAN’s Auto Routing and NAT.

Healthy detection for tunnels

Tunnel Routing maintains a unique mechanism of healthy detection for tunnels, which is different from FortiWAN’s WLHD (See “WAN Link Health Detection”). Symmetric FortiWAN sites continue sending GRE encapsulated detection packets to each other via the defined tunnels. The detection receiver on each FortiWAN site decides the status of a tunnel (OK or Fails) by monitoring if the detection packets arrive continuously. Tunnel Routing’s balancing algorithms distribute packets only over those healthy tunnels, so that the network connection and the data transfer reliability are guaranteed. Tunnel Routing’s healthy detection contains the whole connection between two FortiWAN sites (from the WAN link one side to the WAN link another side via Internet), while WLHD only detects the status of connections to Internet. Therefore, the two mechanisms might show different detection result. For example, the Web UI reports a WAN link is OK but a tunnel established with the WAN link is failed. This might be the failed WAN link on the opposite site of the tunnel. For another example, the Web UI reports a WAN link is failed but a tunnel established with the WAN link is OK. This might because a incorrect configuration to WLHD results in incorrect detection.

Dynamic IP addresses and NAT pass through

FortiWAN’s Tunnel Routing supports dynamic IP addresses and NAT pass through. Only one static public IP address (No NAT employed to the static IP address) is required for tunnel routing deployment between the symmetric FortiWAN sites. A negotiation will be dynamically performed via the only one static public IP address to synchronize the dynamic IP addresses and the IP addresses of NAT device to each other. Therefore, changes on dynamic IP addresses or IP addresses NAT device causes no damage to tunnel connections. Note that NAT pass through for Tunnel Routing here is not the NAT function of FortiWAN, FortiWAN will never perform NAT translation for tunnel packets. The NAT pass through here is for the application that another NAT device in front of FortiWAN. Usually, this happens when a ISP provides WAN links with private IP addresses and does NAT translation for the private WAN links on the ISP side.

Configurations

Auto-routing is a trunking technology that provides load balancing and fault tolerance for all outbound requests, but it does not apply to inbound requests. These are handled by a unique technology called SwiftDNS, a multihoming service which includes load balancing and fault tolerance for inbound requests. The minimum requirements for multihoming are networks must have multiple WAN links and registered domain names for publicly accessible servers. Note that a DNS request from client is delivered to FortiWAN via a fixed WAN link, whose the IP address is registered with parent domain. It would be better to have multiple IP addresses registered to avoid single WAN link failure.

When FortiWAN receives a DNS query, it replies with a public IP assigned to one of the WAN links based on the settings of the answering policies. Therefore, subsequent requests to server will be sent to a public IP of the WAN link based on FortiWAN’s previous response. The policies are based on weight for each WAN link and are definable. Multihoming is also capable of automatically detecting the best links by “Optimum Route”, and if WAN link failure occurs, the public IP assigned to that failed link will not be returned even though the servers are still reachable via other links.

FortiWAN offers two options for Multihoming: Non Relay Mode and Relay Mode. The details of will be explained in this section.

The section explains how to configure Multihoming. First, check the box to enable Multihoming in “Enable Multihoming”. Multihoming supports Backup mechanism. To enable this function, check “Enable Backup” and enter the IP addresses of the backup server.

FortiWAN provides mechanisms to record, notify and analysis on events refer to the Multihoming service, see “Log”, “Statistics: Traffic”, “Statistics: Bandwidth” and “Reports”.

Non-Relay Mode

To enable Multihoming in non-relay mode, go to Service > Multihoming on the Web UI, check the box Enable Multihoming, and uncheck the box Enable Relay. FortiWAN will performs DNS analysis on local host if the relay mode is disabled. It contains three blocks to get non-relay mode Multihoming configured: Global Settings, Policy Settings and Domain Name Settings.

Global Settings: IPv4/IPv6 PTR Record

PTR (Pointer Record) is used to resolve the IPv4/IPv6 address to a domain or hostname.

DNSSEC Support

The DNS Security Extensions (DNSSEC) is a specification that adds data authentications and integrity to standard DNS. To resist tampering with DNS responses, DNSSEC introduces PKI (Public Key Infrastructure) to sign and authenticate DNS resource record sets within the zone. A signed zone includes a collection of new resource records: RRSIG, DNSKEY and DS.

• RRSIG contains the DNSSEC signature for the corresponded DNS records (A, AAAA, MX, CNAME and etc.) within the zone.
• DNSKEY contains the public key corresponded to the private key used to generate RRSIG records. A DNS resolver uses it to verify DNSSEC signatures in RRSIG.
• DS (Delegation Signer) references to the public key used to verify the RRSIG in your zone. Every DS record should be signed by your parent zone and stored in the parent zone to establish trust chain between DNS zones.

Multihoming supports basic DNSSEC which employs only one key pair KSK (Key Sign Key) to generate DNSKEY and RRSIG records for the zone (NSEC is not supported). The supported algorithm and key size are only RSASHA512 and 2048 bits. Note that Multihoming’s DNSSEC is not supported for Relay Mode.

Remember that you have to configure DS records with your domain registrar after you complete configurations for DNSSEC. Please contact your domain registrar for further details about managing DS records.

Relay Mode

For the case that a DNS server already exists in you network, Relay Mode is the way to combine the existing DNS servers with Multihoming’s inbound load balance and fault tolerance. With Relay Mode enabled, FortiWAN will forward all the DNS requests it receives to the specified name servers, in stead of processing the requests directly. Answer of the DNS request will be responded to FortiWAN from the name server. FortiWAN’s

Multihoming then reprocess the answer with appropriate IP address according to the AAAA/A records and AAAA/A policies (load balancing algorithm). The DNS answer that contains appropriate IP address will finally responded to client, so that the inbound access could connect via the appropriate WAN link.

Enable Backup

FortiWAN Multihoming employs Backup mechanism to provide disaster recovery approach for network across various regions. Under this mechanism, the same backup service is set up across different regions. Therefore, when master site is down, backup site will immediately take over to resume the service.

To deploy Multihoming Backup between two FortiWAN units for one domain, at least one of the WAN links’ localhost IPv4 addresses of each FortiWAN unit must be registered with the parent domain (so that a DNS request for the domain can be delivered to the two FortiWAN units). Check “Enable Backup” on the Slave FortiWAN Web UI and specify the IPv4 addresses (which are registered with parent domain) of the Master FortiWAN in “Remote Master Servers”. Configurations for Multihoming Backup deployment is only necessary on the Slave unit, please do not check “Enable Backup” on the Master unit.

Then the Slave unit will detect the state of the Master unit periodically with its built-in Dig tool. The detect packets will be delivered to Master unit via the IP addresses specified on the Slave unit. When the Master’s Multihoming works properly, the Slave’s Multihoming will get into non-active mode (Unit that is in non-active mode will not answer to any DNS request); when the Master’s Multihoming is down, the Slave will get into active mode and take over to resume Multihoming. After takeover, the Slave will continuously detect Master’s state. Once the Master recovers, the Slave will return Multihoming service back to Master and get into non-active mode. This is how the Backup mechanism offers disaster recovery function. DNS database synchronization is not provided for Multihoming Backup deployment, so that DNS database can be maintained individually on the two units for local and remote-backup services. In case that multiple IP addresses of FortiWAN are registered with parent domain (to avoid single WAN links failure), those IP addresses should be configured into the “Server IPv4 Address” field on the Slave unit.

Inbound Load Balancing and Failover (Multihoming)

Multihoming

Multihoming is a technique when external users request any server’s IP address; Multihoming promptly returns DNS response according to the link quality. This provides unmatched availability of bandwidth and load-balances incoming traffic across the multiple ISP lines.

Simultaneously using multiple IP address provided by the ISP connections can result in problems with inbound traffic. For example, if the network is currently using an IP provided by ISP1, and a problem occurs with this ISP, then the inbound query will not be received because the external traffic only knows the IP address provided by ISP1. Also, by using the IP provided ISP1, ISP2 cannot manage the inbound traffic of ISP1. Therefore the concern with multiple ISP links is how to effectively display IP address to the external environment.

Multihoming uses DNS fault-tolerance technique to resolve this problems with the simultaneous use of multiple ISP connections. For example, if the web server for external traffic uses a single ISP connection, then any problems with that connection will affect the network. However, if the DNS periodically assigns different IP addresses provided by different ISP connections, then the external traffic will always have a valid IP to connect to. The actual implementation is assigning a name of different IP, and any query to this name will receive an IP address. As a result, different users can access the web server through different IPs, which is the purpose of Multihoming.

Assuming, there are three WAN links (therefore three different IPs) for the web site of www.example.com, the DNS record has three entries:

www IN A 211.21.10.3 www IN A 63.98.110.123 www IN A 192.136.1.243

All DNS requests to www.example.com will be sent to FortiWAN. Multihoming will constantly measure the health conditions as well as the state of each WAN link and compute the optimal return answer to the DNS queries, defined as the SwiftDNS technology. The SwiftDNS technology will not only ensure fault tolerance for inbound traffic, it also supports powerful and flexible load balancing algorithms as in the Auto Routing mechanism to enable users with heavy web presence to maximize the reliability and efficiency of their web services.

The SwiftDNS Multihoming mechanism requires network administrators to understand the details of the system behavior. The fundamental concept of the DNS mechanism is shown in the next section. A step by step deployment tutorial will also be provided.

Introduction to DNS

DNS server differs from the host file based on name resolution. Host file contains information of IP address mapping information. It is only useful for intranet where the information of host machines is relatively static. Name resolution by DNS server is dynamic because it can adapt to changes easily. The way it works is based on DNS server hierarchy on the Internet. If a DNS server cannot resolve a name (the information is not in its cache), it will ask other DNS servers. There is a protocol on how and where to ask other DNS servers.

A name resolution request may go through a number of DNS servers. When an answer is found, it will be saved in cache so that the same request can be answered immediately without asking other DNS servers again. Each name resolution result saved in cache has a TTL (Time To Live). After the period of TTL, it will be discarded in order to avoid stale information.

The whole internet has a large DNS hierarchy. The top of the hierarchy is called Root. It consists of a set of Root DNS servers coordinated by ICANN. The next level below Root is Top Level Domain (TLD). TLD registration database contains information about top level domains such as CA, COM, EDU, GOV, NET, etc. The next level below TLD is Second Level Domain (such as whitehouse.gov, Microsoft.com, inforamp.net, etc.) followed by Third Level Domain, and so on.

You can apply for domains for your organization. First, go to the Internet’s Network Information Center (InterNIC) to find out if the domain has been registered already. You can also consult the ICANN-accredited registrar database. Second, register the domain with a registrar. You have to provide at least two DNS servers to serve DNS requests. If your registration has been approved, then any DNS request to your domain will be forwarded to the DNS servers you are registered with. For example, xtera.com is registered and InterNIC has put the name “xtera” into the COM DNS servers.

Once the domain is registered, sub-domains can be created. Example: a part or the network can be named “sales.xtera.com”. InterNIC’s approval is not required for creating sub-domains. However, it is important to put DNS information about sales.xtera.com into the DNS servers of xtera.com.

Here is an example of how DNS hierarchy works. A user at a university sees a link to sales.xtera.com on a web page and clicks it. The browser will ask the local DNS server dns.utexas.edu about sales.xtera.com. Suppose it is not in the cache of dns.utexas.edu. The DNS server goes to a Root DNS server to find the DNS server for COM TLD. The DNS server for COM TLD tells dns.utexas.edu to go to dns1.xtera.com. Finally dns.utexas.edu is given the IP address of sales.xtera.com by dns1.xtera.com.

SwiftDNS

One of the problems with traditional DNS servers are facing is TTL. A long TTL means a long update time when IPs have been changed. Before the update time is up (i.e. TTL is expired), DNS requests may be answered with incorrect information. FortiWAN employs SwiftDNS for multihoming based on the health state of the link and a traffic re-directing algorithm. SwiftDNS dynamically answers DNS requests to prevent broken or congested links. In order to solve the TTL issue stated above, SwiftDNS maintains a very short TTL and actively sends out updates to internal DNS in case of link status changes.

How does SwiftDNS work?

Here is an example to illustrate how SwiftDNS works. When Multihoming is enabled, SwiftDNS becomes active. In this case, the upper level DNS server for example.com has two NS records and they are for Primary DNS server at 210.58.100.1 and Secondary DNS server at 210.59.100.1. Both of them are pointing to FortiWAN.

In this case, a web site at 192.168.100.1 in LAN is exposed to these two IPs. When both ISP links are working properly, FortiWAN replies to DNS requests for www.example.com with 210.58.100.1 and 215.59.100.1 at ratio of 1:2 (weight ratio).

Assuming ISP1 is down and a DNS request for www.example.com comes in, it would not be able to go through 210.58.100.1 but it will be able to reach 215.59.100.1. Multihoming detects the link status of WAN1 and answer the request with 215.59.100.1.

Prerequisites for Multihoming

In order to multihome properly, review the requirements below.

Prerequisites for Multihoming:

• Multiple WAN links (minimum of 2).
• Registered domain names for public servers. Please make sure DNS requests for the domains can be delivered to FortiWAN. l Public servers must be configured as virtual servers, or have public IPs

Besides, Multihoming is a non-recursive name server which is an authoritative DNS service that allows others to find your domain only. Multihoming does not answer for unknown domains.