chia-plotter
chia-plotter copied to clipboard
madMAx43v3r
Add options for a reducing amount of temp2 space used at any given time
Related to #819 and #805
By using a RAM block cache as small as 40GB, almost 50% of writes to a temp2 SSD volume can be prevented.
If would be great if this concept could be extended to adjust chia_plot to restrict the amount of data it is processing at any given time in the temp2 dir such that it might completely fit into a user's RAM block cache or RAM disk.
This kind of feature would be very valuable for the Chia community, since it could give users with less system memory the ability to do more plotting in memory without degrading SSD's. It would also help to negate some of the bad press that Chia gets for it's reputation as an "SSD destroyer".
Tu ya configuraste para que con 40GB de Ram funcione? Si es así, me podrías explicar porfavor? mi placa no soporta 128 gb de RAM, pero 64 si es una opción
@nikobart I used PrimoCache on Windows to configure a 40GB L1 cache (Strategy: Write Only, Defer Write period: infinite) for my temp2 SSD volume. This reduced SSD drive writes by about 50%. I believe the Linux equivalent would be to use bcache. My system only has 64GB RAM as well, so this left enough free memory for the Chia plotter and other background tasks
@nikobart Solía PrimoCache en Windows para configurar una caché L1 de 40 GB (Estrategia: Sólo escritura, Defer periodo Comentario: infinito) para el volumen de mi SSD Temp 2. Esta unidad SSD reduce las escrituras en aproximadamente un 50%. Creo que el equivalente de Linux sería usar
bcache. Mi sistema solo tiene 64 GB de RAM, por lo que esto dejó suficiente memoria libre para el trazador Chia y otras tareas en segundo plano.
Muchas Gracias! última duda, la velocidad de trazado se vio afectada?
Nice idea! However, at least for me, be careful with PrimoCache....latest Win10 x64 and after reboot, during restart, had multiple BSODS and had to hard reset.
The equivalent Linux version would be rapiddisk, but it is quite coumbersome to build/install/use (recompiling kernel etc..)
Would be nice to have a guide for Linux with bcache (or eventually LVM too since it supports caching). I'll start with few commands as example:
installing bcache
sudo apt-get update sudo apt-get upgrade sudo apt-get install bcache-tools
creating & mounting ramdisk
sudo mkdir /mnt/ramdisk sudo mount -t tmpfs -o size=2G ramdisk /mnt/ramdisk sudo mount -a
creating bcache device
sudo wipefs -a /dev/ssd -> this will wipe ssd sudo make-bcache -B /dev/ssd -C /mnt/ramdisk --discard --writeback sudo mkfs.ext4 /dev/bcache0
deleting bcache device
sudo umount -v /dev/bcache0 sudo su - -c "echo 1 >/sys/fs/bcache/device_uuid/unregister" sudo su - -c "echo 1 >/sys/block/bcache0/bcache/stop"
the problem im facing is that, after creating the ramdisk, I launch sudo make-bcache and get the following error: Cannot use "/mnt/ramdisk", it is a folder"....Any idea?
PrimoCache results:
- win10 latest
- 1x nvme as tmpdrive & tmpdrive2 (21G, write only, defer time infinite, block size 4k)
- Total write req 1.12TB (host)
- Total disk write 0,99TB
this is about a 11% of saved writes per plot (confirmed also by CrystaldiskMark). Plot time barely affected (48min ubuntu, 51min win10)
the problem im facing is that, after creating the ramdisk, I launch sudo make-bcache and get the following error: Cannot use "/mnt/ramdisk", it is a folder"....Any idea?
You can't use bcache with a tmpfs ramdisk. You need block device. brd module does it.
new test done:
- E:\ tempdrive1 | D:\ tempdrive2
- cache enabled on D:\
- cache info: 25GB, block size 4kb, write only, defer write infinite, mode native, empty data written from cache
- 1.12TB / 1.43TB -> 21.7% saved ssd writes -> about 300GB per plot
- 6% faster plot time compared to ubuntu
PrimoCache results:
- cache enabled on tempdrive2
- cache info: 55GB, block size 4kb, write-read , defer write infinite, mode native, empty data written from cache.
- tempdrive2 total write 991GB / write to disk 394GB -> 594Gb saved ssd writes
- plot time drop from 48 min to 40 min
