| 21 Aug 2023 |
 @elvishjerricco:matrix.org | whoops | 08:40:40 |
| @elvishjerricco:matrix.org | too late :P | 08:40:44 |
| @elvishjerricco:matrix.org | might change that later | 08:40:56 |
 @linus:schreibt.jetzt | from my experience, an all-HDD binary cache is pretty painful | 08:41:48 |
| @elvishjerricco:matrix.org | I didn't realize accessing narinfos was such a burden | 08:42:32 |
| @linus:schreibt.jetzt | but if you have a special device on your zpool and set special_small_blocks you can get your narinfos on the SSD within the pool | 08:42:52 |
| @linus:schreibt.jetzt | with the bonus of getting small nars on the SSD too | 08:43:01 |
| @linus:schreibt.jetzt | In reply to @elvishjerricco:matrix.org
I didn't realize accessing narinfos was such a burden lots of small files access, lots of very random access. Worst performance case for HDDs ^^ | 08:43:38 |
| @elvishjerricco:matrix.org | I do wish I had a special device in this pool. And I have the hardware for it (yay optane). But I've been planning on rebuilding the pool for a while and planned on adding them then | 08:44:02 |
| @elvishjerricco:matrix.org | In reply to @linus:schreibt.jetzt
lots of small files access, lots of very random access. Worst performance case for HDDs ^^ Yea I just didn't realize narinfos would be a significant amount of the IOPS necessary to serve a binary cache | 08:44:59 |
| @linus:schreibt.jetzt | In reply to @elvishjerricco:matrix.org
I do wish I had a special device in this pool. And I have the hardware for it (yay optane). But I've been planning on rebuilding the pool for a while and planned on adding them then why only then? ^^ | 08:45:26 |
| @elvishjerricco:matrix.org | Because I didn't know I'd have a legit use for it this very night :P | 08:45:44 |
| @linus:schreibt.jetzt | :D | 08:45:49 |
| @linus:schreibt.jetzt | though briefly back to the topic of attic's deduplication: I'm using the suggested chunking parameters, and they result in uploads being fairly slow (the bottleneck appears to be CPU-bound stuff in the postgres database) | 08:47:16 |
| @elvishjerricco:matrix.org | I'm not hugely worried about anything like that. More worried about download speed | 08:48:05 |
| @elvishjerricco:matrix.org | Using this on my LAN | 08:48:10 |
| @linus:schreibt.jetzt | download speed is fine, reading the database is fast and reassembling the nars from chunks is trivial | 08:48:37 |
| @elvishjerricco:matrix.org | though, the single threaded perf of this CPU aint great | 08:48:39 |
 Julien | Do we have measures of the total deduplication of an attic cache ? | 08:48:57 |
| @elvishjerricco:matrix.org | In reply to @linus:schreibt.jetzt
why only then? ^^ I just remembered... I've already got the optane in the machine... I could just add it now and rewrite the narinfo files... | 08:49:53 |
|  Christina Sørensen joined the room. | 08:50:44 |
| @linus:schreibt.jetzt | In reply to @julienmalka:matrix.org
Do we have measures of the total deduplication of an attic cache ? attic doesn't currently have a built-in command for it but it should be pretty easy to get from the db | 08:51:14 |
| @linus:schreibt.jetzt | attic=# select sum(file_size) / (select sum(nar_size) from nar) from chunk;
?column?
------------------------
0.20098691256561418637
(1 row)
| 08:52:57 |
| @linus:schreibt.jetzt | so I guess I'm using 20% as much storage for nars as I would be without dedup | 08:53:15 |
| @linus:schreibt.jetzt | *and compression | 08:53:30 |
| @linus:schreibt.jetzt | of course that isn't entirely representative of how much space the cache takes up including file metadata and the database | 08:54:44 |
| @linus:schreibt.jetzt | but I think nars would be the bulk of my binary cache, so that's already a useful number ^^ | 08:55:09 |
| Julien | Given the state of things with AWS hosting it would be interesting to see how much of that dedup we could benefit from in the cache.nixos.org binary cache without asking for demesurate computationnal stress to the server | 08:55:07 |
| @linus:schreibt.jetzt | yeah | 08:55:57 |
| @linus:schreibt.jetzt | I know some people were interested in looking into options there, do you know if anyone's actually doing anything on that front? | 08:56:25 |
