Interesting. I compared his results (DM and DW in table 4.15) with zpaq -m2 -b0, which are both BWT based methods. zpaq does better most of the Silesia corpus (all but sao) and most of the Calgary corpus. Testing on a 2.0 GHz T3200, 2 cores, Ubuntu:
Code:
matt@matt-M-7301U:~/zpaq$ zpaq -b0 -m2 c silesia-m2-b0 silesia/*
Setting max block size for -m1 or -m2 to -b268.435199
[10] silesia/webster 41458703 -> 6545981 (1.2631 bpc)
[2] silesia/mozilla 51220480 -> 16201277 (2.5304 bpc)
[8] silesia/samba 21606400 -> 4039911 (1.4958 bpc)
Creating archive silesia-m2-b0.zpaq
[4] silesia/nci 33553445 -> 1235922 (0.2947 bpc)
[1] silesia/dickens 10192446 -> 2290551 (1.7978 bpc)
[6] silesia/osdb 10085684 -> 2283990 (1.8117 bpc)
[3] silesia/mr 9970564 -> 2147936 (1.7234 bpc)
[12] silesia/x-ray 8474240 -> 3717472 (3.5094 bpc)
[7] silesia/reymont 6627202 -> 1003522 (1.2114 bpc)
[9] silesia/sao 7251944 -> 4730680 (5.2187 bpc)
[5] silesia/ooffice 6152192 -> 2589388 (3.3671 bpc)
[11] silesia/xml 5345280 -> 397102 (0.5943 bpc)
91 seconds