Link Load Balancing – FortiBalancer

Chapter 13 Link Load Balancing (LLB)

13.1 Overview

This chapter details the configuration of the following Inbound and Outbound Link Load Balancing implementations:

  • Single FortiBalancer appliance and two ISPs
  • Dual FortiBalancer appliances and two ISPs

13.2 Understanding LLB

LLB (Link Load Balancing) allows TCP/IP network traffic to be load balanced through up to 128 upstream Internet Service Providers (ISPs). Load balancing can be performed on egress to the Internet (outbound LLB) or on ingress from the Internet (inbound LLB). LLB methods include rr (Round Robin), wrr (Weighted Round Robin), sr (Shortest Response), dd (Dynamic Detecting) and hi (Hash IP). LLB also includes ISP/link failure detection through default gateway and link patch health checking.

The FortiBalancer appliance identifies links based on the logical port and peer MAC address. The statistics of LLB links are also collected based on the logical port and peer MAC address.

13.2.1 Outbound LLB

Outbound LLB provides optimized outbound link utilization for environments that have more than one default gateway. In essence, it allows outbound traffic to be distributed among multiple upstream/ISP routers.

For example, let’s say you have Internet connectivity provided by two ISPs: ISP1 and ISP2. ISP1 assigns address range 100.1.1.0/24 so that you may use them on your network devices. ISP2 assigns address range 200.1.1.0/24 so that you may use them on your network devices. Outbound LLB allows you to load balance outbound connections traffic through ISP1 and ISP2. Connections forwarded through ISP1 are NATTed to an address from the range assigned by ISP1. Connections forwarded through ISP2 are NATTed to an address from the range assigned by ISP2. Thus, inbound responses for those connections will return through the ISP that they were originally sent through. If Internet connectivity through one of the ISP links is lost or interrupted, the outbound traffic will no longer be sent through that ISP. All traffic will be distributed to the functional ISP.

FortiBalancer outbound LLB methods can work well on the data traffic based on TCP and UDP protocols. However, for the packets based on IP, IPsec or GRE protocols, FortiBalancer LLB methods cannot do load balancing well. To deal with this problem, FortiBalancer now supports LLB for the IP based packets so that the IP-based packets can be delivered to different links like the way the TCP/UDP packets are processed.

13.2.2 Inbound LLB

Inbound LLB provides service resiliency for inbound clients. Hosted services are visible to external clients via a separate IP address on the address space assigned by each ISP.

To illustrate, let’s use the same example ISPs as mentioned previously. All external clients trying to connect to the addresses assigned by ISP1 will be routed through ISP1’s backbone. All external clients trying to connect to addresses assigned by ISP2 will be routed through ISP2’s backbone. Inbound LLB allows you to advertise a device or Virtual IP (VIP) using two IP addresses: one from ISP1 and the other from ISP2. A DNS server on the FortiBalancer appliance will respond to queries for configured domain names. The responses will contain an IP address from ISP1 or ISP2, both representing the same device or VIP. If Internet connectivity through one of the ISP links is lost, the DNS server will not respond with the address from the failed ISP. Clients will receive only the address from the functional ISP.

13.2.3 LLB Health Check

LLB Health Check is used to check whether the link between the FortiBalancer interface and the upstream device is available. This can be accomplished by broadcasting ARP requests at regular intervals and pinging a user-defined upstream IP address. Besides, TCP-based and DNS-based health checks are also supported. The ICMP, TCP and DNS types of health check all work in the userland. This greatly improves the health check performance.

Broadcasting ARP requests at regular intervals can check the availability of the link path between FortiBalancer interface and the upstream ISP router. Pinging a user-defined upstream IP address not only can verify if the link path between FortiBalancer interface and the upstream ISP router is available, but also verify the link path between upstream ISP router and user-defined upstream IP address. Multiple upstream IP addresses can be defined for reliable checking. If any of check point is pingable, the related link is usable. This ensures that the WAN link is up before forwarding traffic across that link.

13.2.4 LLB Methods

Outbound LLB supports the following load balancing methods:

  • rr (Round Robin)
  • wrrr (Weighted Round Robin)
  • sr (Shortest Response Time)
  • dd (Dynamic Detecting)
  • hi (Hash IP)

Inbound LLB supports three load balancing methods:

  • rr (Round Robin)
  • wrrr (Weighted Round Robin)
  • proximity

Round Robin distributes each new session to gateways in an alternating (round robin) way. This is the default load balancing method.

Weighted Round Robin is similar to Round Robin except that a bias (or weight) may be assigned to each gateway so that some gateways may receive more sessions than others. This allows more traffic to be directed through an ISP with higher bandwidth capacity.

Shortest Response Time: The link with the shortest response time will get the next request. Calculation of shortest response time of a link is based on the initiation process of each TCP connection (both inbound and outbound connections). For the most accurate result, there should be enough TCP traffic instead of a few long existing TCP connections or only UDP traffic.

Note:

If neither SLB traffic nor NAT traffic goes through the system, the LLB SR method cannot work properly.

The “sr” method cannot be used to load balance IP fragments, non-TCP/UDP packets, and reassembled UDP packets.

Dynamic Detecting performs proximity calculations through all available ISP paths to the destinations. By using parallel probe arithmetic, a request from the client will be sent to a destination by different ISP paths at the same time. When the first response returns, the optimal ISP with the shortest response time will be selected for this request and other ISP connections will be failed. For future outbound traffic to the same destination, FortiBalancer appliance will choose the best ISP connection, according to the results derived from these proximity calculations. Hash IP distributes the outbound traffic among links in the way that the link with higher weight is routed with higher probability, by performing Hash operation on the source IP. When the chosen link is down, the system will carry another Hash operation on the links available. When HI is deployed as the LLB method, the IPflow function can be disabled.

Proximity: The IP address of the nearest DNS server will be sent to the client as the response. When a DNS request arrives, FortiBalancer will first search in the Eroute table reversely to find a proximity route matching the source address of the DNS request, and then give response to the client with the corresponding DNS server’s IP address (A record) according to the Eroute gateway.

13.2.5 Policy-based Routing (Eroute)

LLB policies provide the methods necessary to allow administrators to direct outbound traffic to a preferred route based on the IP address (source and destination) and service type (mail, FTP, Web, etc.).Policy based routing, unlike regular routing, allows the inclusion of the source IP, source port and destination port as well as the protocol into the route selection. For example, using routing policy can ensure that all the traffic generated by AOL instant messenger always uses the same link. If instant messenger client uses different destination IP addresses in its requests and these requests are sent through the different routes, this might confuse the server and cause login failure.

Configuring routing policy will prevent this problem. The CLI command for that would be:

FortiBalancer(config)#ip eroute aol_route 1500 0.0.0.0 0.0.0.0 0 0.0.0.0 0.0.0.0 5190 tcp gateway_ip 1

The FortiBalancer appliance supports at most 5000 eroutes.

IP region

Eroute supports IP region. Administrators are allowed to import pre-defined IP region table via HTTP, FTP or Local File method, and then execute the command “ipregion route” to apply the imported IP region table. This will generate a large number of Eroute configurations, without making complex configurations. Administrators are also allowed to export the IP region table via FTP URL or Local File method.

FortiBalancer appliance will check the contents of the file instead of the file type when an IP region file is imported. To ensure that the IP region file can be imported successfully, please pre-define the file contents strictly with the following items included in each entry:

  • IP subnet (in CIDR format)
  • Country name (optional, up to 7 bytes)
  • Brief description (optional, up to 63 bytes) These items must be separated with a “Tab”. For example:
27.8.0.0/13 CN China Unicom Chongqing Province network
27.36.0.0/14 CN China Unicom Guangdong province network

Note: 

  1. By default, there are three predefined IP region tables including “predefined_cernet”, “predefined_cnc” and “predefined_ct”. It is recommended not to use the same name with the default predefined IP region tables.
  2. The routes and proximity rules configured for IP region exist as a whole in the system.

Administrators cannot change or remove a single route or a rule.

13.2.6 LLB Session Timeout

After an ISP link has been selected for an IP flow (source IP and destination IP) pair, all traffic with the same source IP and destination IP will be sent to the same ISP. After an IP flow has been idle for a period of time, the session will be removed. Subsequent IP flows will once again be distributed based on the load-balancing algorithm.

13.2.7 Route Priority

The administrator will need to provide the method necessary to allow end-users to direct outbound traffic to a preferred route based on the IP address and protocol type. FortiBalancer appliance supports variant types of routing rules in which eroute priority is higher than priority of the default and static routes. Default routes will have priority 1 and static routes 101-132 depending on the netmask; i.e. the static route with 24-bit netmask will have priority 124 and with 32-bit netmask will have priority 132. The routes that correspond to the interfaces will have priority 2000. The routes created based on the traffic that come from the local subnet are called droutes (Direct Route) and will have priority 2000.

The following table shows the priority of different types of routes:

Table 13-1 Route Priority

Name of Route Priority
EROUTE-P 2001-2999
IROUTE, DROUTE 2000
RTS 1999
EROUTE-N 1001-1999
IPFLOW 1000-1999 (defaults to 1000)
STATIC ROUTE 101-132 (IPv4)

101-228 (IPv6)

DYNAMIC ROUTE 101-132 (IPv4)

101-228 (IPv6)

LLB LINK ROUTE 2
DEFAULT ROUTE 1

13.2.8 Link Bandwidth Management

For better link bandwidth management, the FortiBalancer appliance allows administrators to set a threshold value for the LLB link bandwidth.

When performing link selection for the outbound traffic, the system considers not only the routing policies configured for links but also the load status of each link. That is, when the current link has reached the configured bandwidth threshold, the FortiBalancer appliance will search for available links from matched routes according to the descending sequence of priorities. The FortiBalancer appliance first searches for available links from routes with the same priority as the current link. If all available links reach their bandwidth thresholds, the FortiBalancer appliance will search for available links from routes with lower priorities. If the gateways of all matched routes are down or reach the configured bandwidth thresholds, the FortiBalancer appliance will still choose the current link to transmit traffic.

In addition, the FortiBalancer appliance allows administrators to configure a priority for the LLB link bandwidth. If the priority of a matched route is higher than the LLB link bandwidth priority, the traffic will be directly forwarded through this route.

With the LLB bandwidth management function, you do not need to configure Eroutes with the same priorities for multiple links. This improves the efficiency and flexibility of link bandwidth configuration and management.

Note:

  1. If the traffic hits a RTS or IPflow route, the traffic will be directly forwarded through the relevant LLB link no matter whether the LLB link reaches the bandwidth threshold.
  2. If an Eroute has been configured with the source IP address, source mask, source port number, destination IP address, destination mask, and destination port number and these IP addresses and masks are set to 0.0.0.0 and port numbers are set to 0, the FortiBalancer appliance will not search for available links from the matched routes whose priorities are lower than 1000.

13.2.9 IPv6 Support for LLB

The FortiBalancer appliance provides broad IPv6 support for the LLB module, of which the Eroute, inbound and outbound LLB, link health check and IP region can all work in the IPv6 network environment. For the Eroute, the source IP, destination IP, gateway IP and IP region can all be configured with the IPv6 addresses. However, please note that only IPv4 or only IPv6 addresses can be configured in one IP region table. For outbound LLB, only route-based LLB supports IPv6 configurations, while NAT-based LLB does not.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.