The SDM is Switching Database Manager . The SDM is implemented in all versions of Cisco IOS ® software for the Catalyst 2948G-L3 and 4908G-L3 switches.
The Switching Database Manager (SDM) on the Catalyst 2948G-L3 and 4908G-L3 Layer 3 switches manages the Layer 2 and Layer 3 switching information maintained in the ternary CAM (TCAM) used by the Gigabit Ethernet interfaces for forwarding lookups.
The Switching Database Manager (SDM) on the Catalyst 2948G-L3 and 4908G-L3 Layer 3 switches manages the Layer 2 and Layer 3 switching information maintained in the ternary CAM (TCAM) used by the Gigabit Ethernet interfaces for forwarding lookups.
The TCAM used by Gigabit Ethernet interfaces is separate from the dynamic CAM tables used by 10/100 Fast Ethernet interfaces (on the Catalyst 2948G-L3). This dynamic CAM is shared by all protocol entries (IP prefix, IP adjacencies, IPX networks, IPX nodes, and so on) and can store a maximum of 24K total entries. However, the 24K entries in the shared CAM must be consistent with the entries in the TCAM used by the Gigabit Ethernet interfaces.
The TCAM space on the Gigabit Ethernet interfaces consists of 32K entries, where each entry is 32 bits wide. By default, the SDM partitions approximately 24K of the 32K TCAM space at bootup. There are two reasons for not partitioning the entire TCAM space:
On the 2948G-L3, the entries in the shared CAM and the TCAM must be consistent. Therefore, on the Catalyst 2948G-L3, it is not recommended that you allocate the remaining 8K entries for protocol-specific regions, to reduce the likelyhood of inconsistencies between the shared CAM on 10/100 interfaces and the TCAMs on the Gigabit interfaces (such inconsistencies are still possible if the number of entries for a given region in the TCAM exceeds the allocated space while there is still space for the entries in the shared CAM).
Future features on the Catalyst 2948G-L3 and 4908G-L3 (such as future support for data-place access control lists [ACLs] on Gigabit ports) will require some portion of the total TCAM space in order to function. Therefore, we do not recommend that you allocate the remaining 8K entries on either the Catalyst 2948G-L3 or the 4908G-L3 unless you require the additional entries and you do not anticipate using data-plane ACLs in the future. Doing so will reduce the TCAM space available for ACL usage and might require that you reallocate your TCAM space to implement these features.
Unlike the shared CAM used on 10/100 interfaces, the SDM partitions switching information in the TCAM into protocol-specific regions. There are two types of TCAM regions, exact-match and longest-match:
Exact-match regions store entries such as IP adjacency entries, IPX network entries, IPX node entries, and Layer 2 (MAC) entries
Longest-match regions store entries such as IP prefix entries, IP multicast entries, and UDP flooding entries
Entries in the longest-match TCAM regions are grouped into buckets from most-specific (longest) to least-specific (shortest) mask length, where all entries within a bucket share the same mask value and key size. The bucket sizes in each protocol-specific region vary. The SDM increases the size of the buckets dynamically during system operation, moving existing TCAM entries as necessary.
To increase efficiency, the SDM autolearn function saves the bucket-size (mask-length) distribution for these regions so that the bucket-size distribution is preserved across a system reload. This increases system efficiency by reducing the need to move TCAM entries due to dynamic bucket sizing. The autolearn function is enabled by default, but can be disabled if desired. The autolearn function is implemented based on the assumption that the prefix distribution in a network is consistent and will be the same after the system is reloaded.
There are eight protocol-specific TCAM regions controlled by the SDM. Six of these regions are user-configurable
The TCAM space on the Gigabit Ethernet interfaces consists of 32K entries, where each entry is 32 bits wide. By default, the SDM partitions approximately 24K of the 32K TCAM space at bootup. There are two reasons for not partitioning the entire TCAM space:
On the 2948G-L3, the entries in the shared CAM and the TCAM must be consistent. Therefore, on the Catalyst 2948G-L3, it is not recommended that you allocate the remaining 8K entries for protocol-specific regions, to reduce the likelyhood of inconsistencies between the shared CAM on 10/100 interfaces and the TCAMs on the Gigabit interfaces (such inconsistencies are still possible if the number of entries for a given region in the TCAM exceeds the allocated space while there is still space for the entries in the shared CAM).
Future features on the Catalyst 2948G-L3 and 4908G-L3 (such as future support for data-place access control lists [ACLs] on Gigabit ports) will require some portion of the total TCAM space in order to function. Therefore, we do not recommend that you allocate the remaining 8K entries on either the Catalyst 2948G-L3 or the 4908G-L3 unless you require the additional entries and you do not anticipate using data-plane ACLs in the future. Doing so will reduce the TCAM space available for ACL usage and might require that you reallocate your TCAM space to implement these features.
Unlike the shared CAM used on 10/100 interfaces, the SDM partitions switching information in the TCAM into protocol-specific regions. There are two types of TCAM regions, exact-match and longest-match:
Exact-match regions store entries such as IP adjacency entries, IPX network entries, IPX node entries, and Layer 2 (MAC) entries
Longest-match regions store entries such as IP prefix entries, IP multicast entries, and UDP flooding entries
Entries in the longest-match TCAM regions are grouped into buckets from most-specific (longest) to least-specific (shortest) mask length, where all entries within a bucket share the same mask value and key size. The bucket sizes in each protocol-specific region vary. The SDM increases the size of the buckets dynamically during system operation, moving existing TCAM entries as necessary.
To increase efficiency, the SDM autolearn function saves the bucket-size (mask-length) distribution for these regions so that the bucket-size distribution is preserved across a system reload. This increases system efficiency by reducing the need to move TCAM entries due to dynamic bucket sizing. The autolearn function is enabled by default, but can be disabled if desired. The autolearn function is implemented based on the assumption that the prefix distribution in a network is consistent and will be the same after the system is reloaded.
There are eight protocol-specific TCAM regions controlled by the SDM. Six of these regions are user-configurable
