The present invention reduces the number of writes to a main memory to increase useful life of the main memory. To reduce the number of writes to the main memory, data to be written is written to a cache line in a lowest-level cache memory and in a higher-level cache memory(s). If the cache line in the lowest-level cache memory is full, the number of used cache lines in the lowest-level cache reaches a threshold, or there is a need for an empty entry in the lowest-level cache, a processor or a hardware unit compresses content of the cache line and stores the compressed content in the main memory. The present invention also provides LZB algorithm allowing decompression of data from an arbitrary location in compressed data stream with a bound on the number of characters which needs to be processed before a character or string of interest is processed.