Chris Tobey
Contributor
- Joined
- Feb 11, 2014
- Messages
- 114
Hi everyone,
This is not a question, just to share some findings of my own system.
Chassis: Super Micro SYS-6028UX-TR4
Motherboard: Super Micro X10DRU-X
CPU: 2 x Intel Xeon E5-2697 v3
RAM: 256 GB of DDR4 ECC
SAS Controller: 2 x LSI 9211-8i in IT mode
HDDs: 6 x 8 TB WD RED + 2 x 16 TB Seagate Exos
FreeNAS: 11.3-U1
So, in this test I have 32TB of data on a pool consisting of the 2 x 16 TB drives in a stripe configuration. The data is not particularly well-suited to deduplication as it is mostly video files, but I have the system for it, so why not try. I created a RAID-Z2 pool with the 6 x 8 TB drives, enabled deduplication, created a snapshot via the GUI, and then set up a replication task for that snapshot in the GUI, then I waited. In this test I look at the read performance of the mirrored drives to simplify things, and the data is evenly distributed so the read performance can be multiplied by two for the total read performance, which should be equal to the total write performance of the deduplicated RAID-Z2 pool.
Things started off pretty nice, but after 24 hours, they were not so pretty. Reads start out at nearly 150 MB/s per drive (300 MB/s total), but as the deduplication table grows, the read (and therefore write) speeds quickly drop. After ~1h and 1 TB of data replicated, speed is down to ~100 MB/s per drive (200 MB/s total). After ~20h only about 16 TB has been replicated and the speed is down to ~50 MB/s per drive (100 MB/s total). At only half done, I gave up and stopped the test.
So, then I destroyed the RAID-Z2 pool, re-created it, and kept deduplication off. I started the replication task again in the GUI, and waited. After ~20h of constant reads the performance has fluctuated between 100 MB/s and 210 MB/s per drive (200-400 MB/s total), but this is clearly much, much better than with deduplication.
If a system with these specifications cannot take full advantage of deduplication, I do not know what could.
This is not a question, just to share some findings of my own system.
Chassis: Super Micro SYS-6028UX-TR4
Motherboard: Super Micro X10DRU-X
CPU: 2 x Intel Xeon E5-2697 v3
RAM: 256 GB of DDR4 ECC
SAS Controller: 2 x LSI 9211-8i in IT mode
HDDs: 6 x 8 TB WD RED + 2 x 16 TB Seagate Exos
FreeNAS: 11.3-U1
So, in this test I have 32TB of data on a pool consisting of the 2 x 16 TB drives in a stripe configuration. The data is not particularly well-suited to deduplication as it is mostly video files, but I have the system for it, so why not try. I created a RAID-Z2 pool with the 6 x 8 TB drives, enabled deduplication, created a snapshot via the GUI, and then set up a replication task for that snapshot in the GUI, then I waited. In this test I look at the read performance of the mirrored drives to simplify things, and the data is evenly distributed so the read performance can be multiplied by two for the total read performance, which should be equal to the total write performance of the deduplicated RAID-Z2 pool.
Things started off pretty nice, but after 24 hours, they were not so pretty. Reads start out at nearly 150 MB/s per drive (300 MB/s total), but as the deduplication table grows, the read (and therefore write) speeds quickly drop. After ~1h and 1 TB of data replicated, speed is down to ~100 MB/s per drive (200 MB/s total). After ~20h only about 16 TB has been replicated and the speed is down to ~50 MB/s per drive (100 MB/s total). At only half done, I gave up and stopped the test.
So, then I destroyed the RAID-Z2 pool, re-created it, and kept deduplication off. I started the replication task again in the GUI, and waited. After ~20h of constant reads the performance has fluctuated between 100 MB/s and 210 MB/s per drive (200-400 MB/s total), but this is clearly much, much better than with deduplication.
If a system with these specifications cannot take full advantage of deduplication, I do not know what could.
