Architecture for network search engines with fixed latency, high capacity, and high throughput
View Patent ↗An improved architecture for a network search engine (NSE) is disclosed herein as including an interface manager, one or more levels of a splitting engine, an array of data processing units (DPUs), and a cascade block. A method for using the improved NSE architecture to form an efficient pointer entry database is also provided. As described herein, the improved NSE architecture simultaneously provides high speed, search throughput, update rate and capacity, coupled with low power and fixed latency searches for all search key widths.
1. An architecture for a network search engine (NSE) comprising:
one or more levels of a splitting engine configured for dividing a database of prefix entries into multiple sub-databases, each bounded in size between a minimum value and a maximum value; and
an array of data processing units (DPUs) coupled to the splitting engine for storing the multiple sub-databases, wherein the one or more levels of the splitting engine are further configured for forming a hierarchical tree structure of the database, wherein the hierarchical tree structure comprises a plurality of nodes extending between a root node and a plurality of leaf nodes, and wherein each of the plurality of leaf nodes corresponds to one of the multiple sub-databases of prefix entries
wherein the one or more levels of the splitting engine form the hierarchical tree structure by assigning a pointer entry to each of the plurality of nodes within the hierarchical tree structure;
wherein a first portion of the pointer entries are stored within a first level of the splitting engine in a first storage device, wherein a second portion of the pointer entries are stored within a second level of the splitting engine in a second storage device, and wherein a third portion of the pointer entries, overlapping the first and second portions, are stored within the first and the second storage devices; and
wherein each of the pointer entries residing within the first and third portions comprises a number of bits that is dependent on the level of the hierarchical tree structure at which the pointer entries respectfully reside.
2. The architecture of claim 1 , wherein the one or more levels of the splitting engine comprises at least one storage device for storing at least some of the pointer entries.
3. The architecture of claim 2 , wherein the storage device is selected from a group comprising: logic gates and registers, Content Addressable Memory (CAM or TCAM) and Random Access Memory (SRAM or DRAM).
4. The architecture of claim 3 , wherein if the splitting engine comprises only one level, the splitting engine is configured for storing only the pointer entries that are assigned to the plurality of leaf nodes within a single pointer table.
5. The architecture of claim 3 , wherein if the splitting engine comprises more than one level, the splitting engine is configured for storing all of the pointer entries by forming a different pointer table for each set of pointer entries that reside at each level of the hierarchical tree structure.
6. The architecture of claim 5 , wherein each level of the splitting engine comprises the same or a different type of storage device.
7. The architecture of claim 6 , wherein each of the pointer entries residing solely within the second portion comprises a number of bits equal to the level of the hierarchical tree structure at which the pointer entries reside minus a number of bits contributed to a parent pointer entry residing within the third portion.
8. The architecture of claim 7 , further comprising an interface manager configured for translating a search instruction into a search key and sending the search key to the splitting engine, which responds by sending the search key and a corresponding pointer entry to the array of data processing units.