1 |
On Tuesday 22 March 2005 12:45, Daniel Drake wrote: |
2 |
> So on every sync, you have to download the entire 260mb ISO file? |
3 |
> |
4 |
> I don't think our mirrors would be very happy about that. |
5 |
> |
6 |
> Daniel |
7 |
> -- |
8 |
> gentoo-dev@g.o mailing list |
9 |
|
10 |
You don't seem to have understanded how zsync works. |
11 |
|
12 |
Suppose that I, as a user, already have yesterday's portage ISO file. |
13 |
And suppose that today, there's about 30 new or updated ebuilds. |
14 |
Also suppose that those ebuilds amount to something like 500 KB. |
15 |
|
16 |
In those conditions, if I update my ISO file today using zsync, I would only |
17 |
have to download the zsync file (which would be about 700 KB) and the |
18 |
necessary *compressed* ranges of the (compressed) ISO file available on the |
19 |
mirror. This would be *less* than 500 KB, because of the compression. |
20 |
|
21 |
This works because the .zsync file contains a mapping of the uncompressed data |
22 |
to the compressed data. |
23 |
|
24 |
Notice that even if the user doesn't have the ISO file yet, he would only have |
25 |
to download about 27 MB. |
26 |
|
27 |
Personally, I estimate that updates could be faster than a rsync, if not only |
28 |
because of the whole disk thrashing. But only through experimentation we |
29 |
would be able to measure the difference. |
30 |
|
31 |
Also notice that zsync still has lots of room for improvements, so I wouldn't |
32 |
be surprised to see it beat rsync in terms of time of an update. |
33 |
|
34 |
I think it's worthwhile to setup an experimental mirror, it sure seems much |
35 |
better than doing emerge-webrsync.. |
36 |
-- |
37 |
gentoo-dev@g.o mailing list |