mjhjmtu

I have output of zdb -DDD geek1 command:

[root@zfs-test centos]# zdb -DDD geek1
DDT-sha256-zap-duplicate: 8193 entries, size 290 on disk, 141 in core

bucket allocated referenced
______ ______________________________ ______________________________
refcnt blocks LSIZE PSIZE DSIZE blocks LSIZE PSIZE DSIZE
------ ------ ----- ----- ----- ------ ----- ----- -----
 2 8K 1G 1G 1023M 24K 3G 3G 3.00G
 16K 1 128K 128K 128K 16K 2G 2G 2.00G


DDT histogram (aggregated over all DDTs):

bucket allocated referenced
______ ______________________________ ______________________________
refcnt blocks LSIZE PSIZE DSIZE blocks LSIZE PSIZE DSIZE
------ ------ ----- ----- ----- ------ ----- ----- -----
 2 8K 1G 1G 1023M 24K 3G 3G 3.00G
 16K 1 128K 128K 128K 16K 2G 2G 2.00G
 Total 8.00K 1.00G 1.00G 1023M 40K 5G 5G 5.00G

dedup = 5.00, compress = 1.00, copies = 1.00, dedup * compress / copies 
= 5.00 

I read about ZFS on internet and also here.
I created 2 files. First is random file from /dev/urandom and second one from /dev/zero. Each file have 1GB. I copied 3x urandom file and 2x zero file to my ZFS server. Minimum block size of ZFS is 128k.

I have following questions:

1. What does refcnt means?


2. What does DSIZE mean in allocated column and how can I count it?


3. There was 8193 entries (unique blocks). What does numbers 290 and 141 means? According this article, I should be able to count how much data table takes in my RAM (8193x141B=1,1MB). If it is ok, what does 8193x290B=2,26MB means? Is that table also saved on disk?


4. What compres and copies in last line means?

Please, include some example if it is possible.

1: refcnt means reference count, i.e. the number of times blocks are referenced by objects (It is a power of two progression so 1 is 1 but 2 actually means 2 or 3¹; 4 means 4,5,6, or 7 and so on)

2: DSIZE= On disk size, LSIZE = logical (in memory), PSIZE = physical. On disk and physical sizes are equal or very close together unless you set zfs to have more than one copy of data blocks in which case on disk size will be larger.

3: 290 bytes is the size a deduplication table entry uses on disk and 141 bytes is the amount of RAM an entry uses. The DDT is stored and synced on disk and indeed currently uses 2.26MB in your case.

4: Compress means the zpool compression ratio (it is unrelated to deduplication). As compression is likely not enabled, the ratio is 1., i.e. no compression at all. Copies means the ratio of ditto blocks (kind of the opposite of deduplication) stored. There are no ditto blocks so the ratio is 1. too.

Example of output with compression and ditto blocks enabled:

bucket allocated referenced
______ ______________________________ ______________________________
refcnt blocks LSIZE PSIZE DSIZE blocks LSIZE PSIZE DSIZE
------ ------ ----- ----- ----- ------ ----- ----- -----
 2 2 256K 49K 98K 6 768K 147K 294K
 Total 2 256K 49K 98K 6 768K 147K 294K

dedup = 3.00, compress = 5.22, copies = 2.00, dedup * compress / copies = 7.84

¹ In your case, the count is clearly exactly equal to 3 (3 GB referenced stored in 1 GB of disk).