Friday, August 23, 2013

deduplication mystery

One of the great things about ZFS (Zeta File System, not zFS, the file system running on z/OS mainframes), as I thought, was de-duplication, i.e. not storing the same file twice.
On the single core 1.8GHz Atom system, the low power processor rendered this feature not feasible. OK, that was to be expected.
However, on the 6 core 3.3GHz build, I figured it would be achievable to run this option. Nope! I tried, before RTFM, and the performance was terrible!
Now that I did RTFM, the dedup-option seems ridiculous. Why? Very simple, it is said that you need at least 16GB of RAM for 1TB of storage when using dedup. I am running a small 3TB server (4 1TB disks in RAIDZ-1). The minimum RAM requirements for using dedub would therefore be 48GB. I believe that this is a little too much for a rather moderate NAS, wondering what they use at data centres.
The wildest server boards have 16 memory slots, taking it to the extremes, one may fill those with 16GB modules, ending at 256GB, which would allow for 16TB of deduplicated storage. For my taste, that's a bit over the moon.
Conclusion, as much as I have liked to employ the dedup-feature in my NAS, I will pass on this. My high performance NAS runs on 8GB only, which I believe is quite something for a mere storage device.