Re: bohe and delta experimentation...

[Mark Nottingham <[e-mail]>]

On 17/01/2013, at 10:20 AM, Nico Williams <[e-mail]> wrote:

> The question is: is it worth the trouble?  I don't have an answer to
> that, but I'm inclined to say that yes, we should represent dates/time
> in an already very small format.

I definitely think it's worth consideration. I just put a flag (compress_dates; you have to set it in source, at least for now) on the simple compressor that lets you see the difference between compression and none...

* Without date compression:

732 req messages processed
              compressed | ratio min   max   std
req  http1       195,386 | 1.00  1.00  1.00  0.00
req simple        82,768 | 0.42  0.16  0.87  0.18

732 res messages processed
              compressed | ratio min   max   std
res  http1       159,968 | 1.00  1.00  1.00  0.00
res simple        92,188 | 0.58  0.12  0.89  0.16

* With date compression:

732 req messages processed
              compressed | ratio min   max   std
req  http1       195,386 | 1.00  1.00  1.00  0.00
req simple        82,357 | 0.42  0.16  0.87  0.18

732 res messages processed
              compressed | ratio min   max   std
res  http1       159,968 | 1.00  1.00  1.00  0.00
res simple        79,930 | 0.50  0.13  0.87  0.15

Note that the traces we currently have don't have If-Modified-Since in them very much, since they were started with a clean cache; this is why there's little difference in requests (Date doesn't occur much in reqs). 

Of course, we can get to the similar results by many other means...

Cheers,


--
Mark Nottingham   http://www.mnot.net/