I think I solved a major problem in the NAS industry
| cock of zohran mamdani | 09/10/25 | | VoteRepublican | 09/10/25 | | cock of zohran mamdani | 09/10/25 | | cock of zohran mamdani | 09/10/25 | | daniel gay luis | 09/10/25 | | Talk to your kids about time dilation. | 09/10/25 | | cock of zohran mamdani | 09/10/25 |
Poast new message in this thread
Date: September 10th, 2025 12:13 AM Author: cock of zohran mamdani
The problem is "write amplification" in NAS systems that were designed to run entirely on hard drives. When running Docker they write tiny amounts of data to the SSD every 5-10 seconds. Normally these would go to hard drives that are always spinning, but over time these writes can add up and shorten the lifespan of a SSD.
My previous approach was to put shitty old SSDs in the cache pools so that I wouldn't care if they died, but then I realized a bunch of these are the old school ones that have a DRAM cache. The DRAM cache shouldn't write to the drive until it's full, and the smallest drive gives you 64mb. The chances of it ever filling up 64mb with these tiny writes is slim, because whatever the OS writes every 5 seconds probably overwrites whatever it wrote 5 seconds earlier.
And now I have proof:
https://i.imgur.com/AUxssNC.png
Two shitty old drives with DRAM caches report a 1:10 ratio of writes to reads. Previously it was 1:3-ish. The cache is usually a ratio of drive capacity, so a 512gb SSD has 128mb of cache, and it's hardly EVER touching that now.
(http://www.autoadmit.com/thread.php?thread_id=5771553&forum_id=2#49249437) |
|
|