| 1 |
On Fri, Jan 17, 2014 at 8:27 AM, Michał Górny <mgorny@g.o> wrote: |
| 2 |
|
| 3 |
> Hello, all. |
| 4 |
> |
| 5 |
> I'm using squashfs to hold my Gentoo repositories on all of my systems |
| 6 |
> for some time. As you probably know, this allows me to save space while |
| 7 |
> keeping portage fast. However, it makes updating the tree quite |
| 8 |
> burdensome and time-consuming. |
| 9 |
> |
| 10 |
> We're already hosting daily gx86 tarballs on our mirrors, and deltas |
| 11 |
> made using diffball. Those can be used with Zac's emerge-delta-webrsync |
| 12 |
> to get daily updates done with minimal network overhead. Sadly, it |
| 13 |
> takes the whole process even more time consuming :). |
| 14 |
> |
| 15 |
> Therefore, I'd like to suggest an alternative solution that could help |
| 16 |
> out Gentoo users that use squashfs for gx86 and would like to be able |
| 17 |
> to get daily updates fast and easy. |
| 18 |
> |
| 19 |
> The idea is to host -- along with the tarballs -- daily squashfs images |
| 20 |
> of gx86 in a chosen format. Additionally, the images would come with |
| 21 |
> deltas made using xdelta3 or a similar tool. Those deltas -- with |
| 22 |
> a slight download overhead -- would allow very fast updates |
| 23 |
> of the squashfs. |
| 24 |
> |
| 25 |
> Now some numbers. I did some tests 'converting' late gx86 daily |
| 26 |
> tarballs to squashfs. I've used squashfs 4.2 with LZO compression |
| 27 |
> since it's quite good and very fast. |
| 28 |
> |
| 29 |
> 96M portage-20140108.sqfs |
| 30 |
> 96M portage-20140109.sqfs |
| 31 |
> 96M portage-20140110.sqfs |
| 32 |
> 96M portage-20140111.sqfs |
| 33 |
> 96M portage-20140112.sqfs |
| 34 |
> 96M portage-20140113.sqfs |
| 35 |
> 97M portage-20140114.sqfs |
| 36 |
> 97M portage-20140115.sqfs |
| 37 |
> |
| 38 |
> For deltas, I've used xdelta3 with max compression (-9) and djw |
| 39 |
> secondary compression (it gave ~0.1M smaller files than fgk |
| 40 |
> and ~0.5M gain than with no secondary compression). |
| 41 |
> |
| 42 |
> 4,9M portage-20140108.sqfs-portage-20140109.sqfs.vcdiff.djw |
| 43 |
> 6,3M portage-20140109.sqfs-portage-20140110.sqfs.vcdiff.djw |
| 44 |
> 5,6M portage-20140110.sqfs-portage-20140111.sqfs.vcdiff.djw |
| 45 |
> 8,9M portage-20140111.sqfs-portage-20140112.sqfs.vcdiff.djw |
| 46 |
> 6,3M portage-20140112.sqfs-portage-20140113.sqfs.vcdiff.djw |
| 47 |
> 7,8M portage-20140113.sqfs-portage-20140114.sqfs.vcdiff.djw |
| 48 |
> 8,5M portage-20140114.sqfs-portage-20140115.sqfs.vcdiff.djw |
| 49 |
> |
| 50 |
> As you can see, the deltas are quite large compared to the actual |
| 51 |
> changes. However, we could have expected that since we're diffing |
| 52 |
> a compressed filesystem. What's important, however, is that applying |
| 53 |
> it takes ~2.5 second on my 2 GHz Athlon64. |
| 54 |
> |
| 55 |
|
| 56 |
It wasn't clear to me, are these trees with metadata included? |
| 57 |
|
| 58 |
-A |
| 59 |
|
| 60 |
|
| 61 |
> |
| 62 |
> So, even with the extra download time, the update is much faster |
| 63 |
> than recreating the squashfs. And unlike some types of unionfs, |
| 64 |
> it doesn't come with extra runtime slowdown. |
| 65 |
> |
| 66 |
> What do you think? |
| 67 |
> |
| 68 |
> -- |
| 69 |
> Best regards, |
| 70 |
> Michał Górny |
| 71 |
> |
Archives