meta data for this page
  •  

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
vm:proxmox:pbs:performance [2023/04/25 12:30] – niziakvm:proxmox:pbs:performance [2026/03/20 07:13] (current) – niziak
Line 1: Line 1:
 ====== performance ====== ====== performance ======
 +
 +===== FS tuning =====
 +
 +PBS stores everything in files. Maximum file size is 4MB.
 +  * Block device backup is in 4MB chunks
 +  * File backup chunk varies from 64kB to 4MB
 +
 +Filesystems:
 +  * ZFS 
 +    * [[https://pbs.proxmox.com/docs/sysadmin.html#zfs-special-device|special device]]
 +    * 128kB recordsize can cause write amplification for chinks smaller than 128kB
 +    * larger recordsize = less IOPS
 +    * Create histogram from existing PBS chunks sizes.
 +
  
 <code> <code>
Line 7: Line 21:
 read/written from the HDDs). read/written from the HDDs).
  
-In general HDDs shouldn't be used with PBS, atleast not if you store alot of backups. And if you +In general HDDs shouldn't be used with PBS, at least not if you store a lot of backups. And if you 
 still do it, it's highly recommended to also use SSDs for storing the metadata. still do it, it's highly recommended to also use SSDs for storing the metadata.
 </code> </code>
Line 17: Line 31:
 So:  So: 
   * RAIDZ on HDDS is very slow (due to IO OPS limit to single HDD). Throughput is n x HDD, but IOOPS equals single HDD   * RAIDZ on HDDS is very slow (due to IO OPS limit to single HDD). Throughput is n x HDD, but IOOPS equals single HDD
-  * For HDDs use stripped mirrors to multiply IOOPS (i.e. for 6 disk: 3 stripes per 2 hdd in mirror)+  * For HDDs use stripped mirrors to multiply IOOPS (i.e. for 6 disk: 3 stripes per 2 x HDD in mirror) 
 +  * GC on HDD can takes days without metadata caching on SSD 
 +  * Use SSH as cache: 
 +    * MFU only or 
 +    * configure to cache metadata only
   * Use DRAID - where groups of stripes are used by design.   * Use DRAID - where groups of stripes are used by design.
-  * Use at least separate ZFS dataset for backup, then set block size to <code bash>zfs set recordsize=1M YourPoolName/DatasetUsedAsDatastore</code>+  * Use at least separate ZFS dataset for backup, then set block size to <code bash>zfs set recordsize=16M YourPoolName/DatasetUsedAsDatastore</code> 
 +  * Disable atime <code bash>zfs set atime=off backup2</code> 
 +  * Compression is performed on client side. So enable light compression <code bash>zfs set compression=lz4 YourPoolName/DatasetUsedAsDatastore</code>