Some SSD write volumes from my machines

February 8, 2022

A while back I read Russell Coker's SSD Endurance (via Planet Debian), which got me curious about how much data I had written to my own SSDs (and then led to the discovery of things like odd values for the SMART 'power on hours' attribute). So here is some data drawn from an assortment of machines.

My home desktop has a pair of Crucial MX 750s that have been powered on for 43,550 hours or almost five years, which is actually a plausible number for these drives. They each report about the same number for 'Total LBAs written' at about 79,212,678,300. I believe these are 512-byte units, which would make that about 36.9 Tbytes written over the lifetime of these drives so far, or to put it another way, about 888 Mbytes an hour (or about 20.8 Gbytes a day). These disks are used for my ext4 software RAID root filesystem, my mirrored swap partition, and a ZFS pool that holds both my home directory and various chunks of source code, including Go and Firefox (both of which I build from source on a reasonably regular basis). 888 Mbytes an hour feels high, but who knows.

(I have Prometheus metrics for my home desktop and the numbers come out at least plausible for the write volume. Prometheus says it's actually a bit above 20 Gbytes/day these days.)

My work desktop has a pair of NVMe drives that are used in the same way as my home SSDs and a pair of SSDs that are used exclusively for a ZFS pool, both of them dating from the end of 2019 (as covered in my work desktop's current partitioning). All of the drives have a bit over 19,000 power on hours (the NVMe drives have a few more days than the SSDs). In that time, the NVMe drives have written around 5.5 Tbytes total or (if I'm doing the math right) about 270 Mbytes an hour. One possible reason for the much lower write volume compared to my home machine is that I've only very rarely been in the office since March 2020. My work desktop's SSDs report write volume as 'host writes GiB', with a total of about 22 Tbytes over the time, or the surprising value of 1.17 Gbytes an hour or so. My Prometheus metrics say that host level recorded writes over the past 90 days are about 2.0 Tbytes, or under a Gbyte an hour.

Unfortunately, I seem to be coming up short for interesting servers of ours that have trustworthy SMART data for drives that had been active for a reasonable amount of time. Either the machines are boring (or too excitingly un-representative), or the drives are too young (sometimes implausibly so).

(I'm somewhat reluctant to trust SMART information on this from our fileservers, but on a quick look the drive write volume can vary substantially between disk to disk even on the same fileserver. This isn't surprising, since the utilization of the various different ZFS pools we have varies and clearly the amount that people write to them varies at least as much.)

Written on 08 February 2022.
« What does it mean for a filesystem to perform well on modern hardware?
ZFS performance and modern solid state disk systems »

Page tools: View Source, Add Comment.
Login: Password:
Atom Syndication: Recent Comments.

Last modified: Tue Feb 8 22:04:20 2022
This dinky wiki is brought to you by the Insane Hackers Guild, Python sub-branch.