- VRRP over Ethernet
- VRRP and Learning Bridges
- VRRP over FDDI
- VRRP over Token Ring
- VRRP over ATM LANE
- Summary
This chapter is from the book
This chapter is from the book
This chapter is from the book
5.4 VRRP over Token Ring
Token Ring has several characteristics that make running VRRP over it difficult.
5.4.1 The Token Ring Problem
Chiefly, there is no generally available multicast mechanism available over old and new implementations of Token Ring. The newer implementations of Token Ring do support multicasting using group addresses, but Token Ring functional address support is the only generally available multicast mechanism for Token Ring. Also, only a limited number of functional addresses are available, and these could collide with use of the same functional address for another purpose.
There are 31 Token Ring functional addresses, each 48 bits long, organized into 17 bits of constant followed by 31 bits, with only 1 bit on to denote multicast. We specifically chose to discuss the Token Ring addresses in both canonical and noncanonical formats. We did this because the authors of the VRRP RFC draft chose to define functional addresses in canonical (normal) format, and we would like to reconcile that with the Token Ring format here for the benefit of the draft readers. In canonical format, the Token Ring address must be read with each octet flipped (the leftmost bit within an octet is the lowest-order bit and the rightmost bit is the higher-order bit). For example, the Token Ring functional address begins with 03 Hex, or 0000 0011 binary, which means the higher-order two bits are set to 1 in the first octet for a Token Ring functional address rather than the lower-order bits as with Ethernet. This scheme is used because the Token Ring standard requires that addresses supplied to or received from Token Ring interfaces are usually laid out in memory, with the bits of each octet in the opposite order from that of Ethernet, that is, with bit 0 in the high-order (leftmost) position within the octet.
The Token Ring functional addresses start with 03:00, and the addresses range from 03:00:00:00:00:80 to 03:00:02:00:00:00. So 03:00:00:00:00:80 would be the first, followed by 03:00:00:00:00:40, followed by 03:00:00:00:00:20, 03:00:00:00: 00:10, etc. In noncanonical format, the Token Ring address range can be re-written as c0:00:00:00:00:01 through c0:00:40:00:00:00.
Out of these 31 functional addresses, the Token Ring architecture reserves only 12 functional addresses, ranging from 03:00:00:10:00:00 through 03:00:02:00:00:00 for user-defined applications (or in noncanonical format c0:00:00:08:00:00 through c0:00:40:00:00:00). Therefore, any protocol configured to use a multicast address must compete for one of these 12 functional addresses. Additionally, the Novell IPX protocol uses the functional address 03:00:00:10:00:00 over Token Ring, leaving 11 functional addresses that could be used by all other protocols including VRRP. In general, Token Ring VRRP users must take care to avoid functional address conflicts and assign an appropriate functional address to VRRP and their other applications.
5.4.2 The Token Ring Solution
Due to this aforementioned problem, the preferred mode of operation over Token Ring will be to use a Token Ring functional address for the VRID virtual MAC address. The VRIDs are directly mapped to Token Ring functional addresses. To minimize the likelihood of functional address conflicts, the draft standard suggests an allocation that begins with the largest functional address and recommends that VRRP choose VRIDs sequentially starting with 1. This allows for 11 virtual routers at any given time, as shown in Table 5-1.
Table 5-1. VRRP Functional Addresses over Token Ring
|
VRID |
Token Ring Functional Address (Canonical) |
Token Ring Functional Address (Noncanonical) |
|
1 |
03:00:02:00:00:00 |
C0:00:40:00:00:00 |
|
2 |
03:00:04:00:00:00 |
C0:00:20:00:00:00 |
|
3 |
03:00:08:00:00:00 |
C0:00:10:00:00:00 |
|
4 |
03:00:10:00:00:00 |
C0:00:08:00:00:00 |
|
5 |
03:00:20:00:00:00 |
C0:00:04:00:00:00 |
|
6 |
03:00:40:00:00:00 |
C0:00:02:00:00:00 |
|
7 |
03:00:80:00:00:00 |
C0:00:01:00:00:00 |
|
8 |
03:00:00:01:00:00 |
C0:00:00:80:00:00 |
|
9 |
03:00:00:02:00:00 |
C0:00:00:40:00:00 |
|
10 |
03:00:00:04:00:00 |
C0:00:00:20:00:00 |
|
11 |
03:00:00:08:00:00 |
C0:00:00:10:00:00 |
Table 5-1 is generated by mapping VRID to functional addresses via the following function in noncanonical format (0x4000 >> (VRID – 1)) for the middle two octets of a functional address. Programmers may be familiar with the use of ">>" as the bit right shift operator. The above statement means taking the VRID, subtracting it by 1, and then applying right shift (VRID – 1) on 0x4000. To illustrate, let us consider a virtual router with VRID 5. 0x4000 can be written in binary as follows: 0100 0000 0000 0000. Applying right shift on this by (VRID – 1) means (5 – 1) = 4. This means right-shift the 1 by 4 locations, yielding 0000 0100 0000 0000. Converting back to hexadecimal yields 0x0400. Now insert this value into the middle two octets of a noncanonical format of the functional address. For example, if we assume a function address format of C0:00:xx:yy:00:00, then we insert the value of 04:00 into locations xx:yy respectively to yield C0:00:04:00:00:00 for VRID 5. Similarly for VRID 8, right-shift of 0x0400 by 7 times would yield 0x0080, which can be inserted into the xx:yy locations to yield C0:00:00:80:00:00.
5.4.3 Problems with Functional Addresses and Workarounds
However, using functional addresses has problems of its own.
The Host to Virtual Router Forwarding Issue
When VRRP runs in a multiple-ring source route bridging environment as shown in Figure 5-6 and the VRRP routers (master and backup) are on different rings, using functional addresses can cause confusion with respect to the Routing Information Field (RIF) when a host is attempting to talk to the virtual router. For example, current master router R1 is on ring 1, and current backup router R2 is on ring 2. Hosts H1 and H2 are on ring 1. The backup router R2 is on ring 2. If R1 fails, R2 will become the master. Now hosts H1 and H2 will need a RIF to send packets over to ring 2 to the new master router R2. But because the router that becomes the active router, R2, is using the same functional address as that of R1, the stations have no way of knowing that they need to send explorers1 to get the new RIF. Therefore it is better to put both active and standby routers on the same ring/bridge segment.
)
Figure 5-6. VRRP in a multiple-ring source route bridging
environment
The VRRP Advertisement Issue
Functional addresses cannot be used as a MAC-level source address in any packet. Therefore, the VRRP advertisements over Token Ring cannot be sent with the VRRP virtual MAC address as a source address, as is the norm with Ethernet.
Instead the real MAC address, also called the burned-in MAC address, is used in the VRRP advertisements for the source address, while the destination address is set to the VRID functional address. Some older implementations might also use the all-rings broadcast address for the destination MAC address in VRRP advertisements if they cannot support functional addresses, which is inefficient. The VRRP RFC defers to RFC 1469 for what destination address to use in VRRP advertisements; this RFC recommends (but does not mandate) using C0-00-00-04-00-00 as the destination MAC address for all IP multicast traffic. However, because of shortage of functional addresses in general, the destination MAC address is highly configurable, and there is no harm in using the VRID functional address as the destination address in VRRP advertisements as well. Note here that only the VRRP advertisements use the real MAC address as the source. All gratuitous ARPs still use the functional address in the source MAC field. Therefore, from a host's perspective there is still one virtual MAC address, namely, the functional address.
Using the real MAC address as source has its own disadvantage: Proxy ARP cannot function anymore, because a standby router with a different MAC address cannot cover for the lost proxy ARP database of the failed router.
5.4.4 Unicast Mode Operation
Newer Token Ring adapter implementations support nonpromiscuous reception for multiple unicast MAC addresses like Ethernet and thereby avoid both the multicast traffic and the usage conflicts associated with using Token Ring functional addresses. This means that instead of using a multicast mechanism such as functional addresses for the VRRP MAC address, more than one host in a virtual router group can simply register for the standard unicast VRRP MAC address 00:00:5E:00:01:VRID.
However, one important difference exists between unicast mode operation over Token Ring and Ethernet. As Appendix A explains, the ARP request/reply packets from a virtual router on Ethernet have their own MAC address (and need not have the VRRP VRID MAC) in the layer 2 header of the packet. Only the ARP header needs to contain the VRRP VRID MAC in the request or reply. In a unicast mode of operation over Token Ring, even the layer 2 header of an ARP request/reply packet must use the VRRP VRID MAC as the source MAC address. This is necessary because some Token Ring driver implementations keep a cache of MAC address and source routing information independent of the ARP cache, and these implementations will not transmit to a MAC address in a source-route bridged network unless they receive a packet with the same source MAC address in the Token Ring header.
Unicast mode on Token Ring has one limitation similar to the problem outlined in Figure 5-6. If the master and backup routers are on two different source-route bridge segments and there are some host implementations that keep their source-route information in the ARP cache, do not listen to gratuitous ARPs, and listen only to ARPs that they originate, these hosts will not update their ARP source-route information correctly when a failover occurs. As in the previous problem, the only solution is to put all routers with the same VRID on the same source-bridge segment and then use high-availability techniques to prevent that bridge segment from being a single point of failure.