Re: Choosing a header compression algorithm

[Roberto Peon <[e-mail]>]

I've checked in some changes to delta2 which expands and documents various
options for delta2 in the README.md.

After running a number of variations of delta2, The following defaults look
good for small buffer sizes:

delta2=max_entries=256, small_index=1

small_index basically says use a uint8 instead of a uint16 for representing
indices, and is the kind of thing that could be messaged somewhere (opcode,
flag, whatever).

The best headerdiff option which I believe is safe against CRIME in the
future is:
  headerdiff=delta_type=false,huffman

I removed prefix matching from delta some months ago (~6 I think?) after
cogitating on it for a while and then speaking with security folks.. I just
couldn't come up with a way I could prove was safe, unlike the
atom-matching, which one can prove is no worse than a brute-force attack.

I've appended runs with these values@4k buffer size for delta2 and
headerdiff below.
-=R



* TOTAL: 5949 req messages

                                          size  time | ratio min   max   std

                           http1     3,460,925  0.13 | 1.00  1.00  1.00
 0.00
  delta2 (max_byte_size=4096, max_entries=256, small_index=1, hg_adjust=0,
implicit_hg_add=0, refcnt_vals=0)       664,683  4.16 | 0.19  0.02  0.83
 0.15
                                                         headerdiff
(buffer=4096, delta_type=false, huffman)       759,783  2.03 | 0.22  0.01
 0.78  0.18

* TOTAL: 5948 res messages

                                          size  time | ratio min   max   std

                           http1     2,186,162  0.12 | 1.00  1.00  1.00
 0.00
  delta2 (max_byte_size=4096, max_entries=256, small_index=1, hg_adjust=0,
implicit_hg_add=0, refcnt_vals=0)       585,475  5.32 | 0.27  0.02  1.28
 0.13
                                                         headerdiff
(buffer=4096, delta_type=false, huffman)       543,047  3.29 | 0.25  0.02
 0.73  0.14