Only a little OT: SSDs in server farms

riredale wrote on 2/26/2016, 3:09 PM

Google has released a paper discussing the reliability of SSDs versus mechanical hard drives. The results are based on "millions of drive bays over 6 years" and includes a variety of hard drives and SSD technologies. This article on ZDnet takes a first look at that paper.

(1) SSDs wear out over time, not due to use. So they are a bit like lithium batteries in the sense that there is a service life, regardless of usage.

(2) Different technology SSDs are about the same in reliability.

(3) Consumer hard drives are about as reliable as specialized drives.

(4) Enterprise SSDs are more expensive because of over-provisioning (bad block replacement) but that over-provisioning is not really needed.

(5) Uncorrectable bit errors are higher than with hard drives. The writer says it's important to do backups.

Comments

VideoFreq wrote on 2/26/2016, 4:27 PM

Good to know. I think a lot of us have learned this the hard way. SSD's are like a bad race horse, fast out of the gate but slower in the stretch. I use an SSD for my OS and programs but all else is on 7200 RPM HDD's. I make backups of the SSD on Acronis with every new program added.

One thing to note, when I had trouble with my Samsung 840 EVO slowing, I ran across their Magician utility that restores speed but also lets you test your other drives and baseline their speeds. I was quite surprised that across the same exact Seagate 1TB Barracuda drives that the sequential read/write and random (IOPS) read/write speeds were quite different.

GeeBax wrote on 2/26/2016, 4:32 PM

I regard all drives as consumables, no matter whether mechanical or solid state. I give them a nominal 4 year life and replace them at that time. As a result, I have not had a disk crash in as long as I can remember.

I did have a batch of fault SSD, made by OCZ, but so did everyone who used that brand. However they were not used in a critical system, and I caught them and weeded them out early in the piece.