Linux Solid-State Drive Benchmarks
Phoronix: Linux Solid-State Drive Benchmarks
With the number of netbooks on the market continuing to increase each month and more of these mobile devices switching to solid-state drives for their reliability, extended battery life, and faster performance, SSDs are becoming quite common and finding themselves meeting many Linux hosts. How though does the real-world performance differ between hard disk drives and solid-state drives on Linux? We have run several tests atop Ubuntu on a Samsung netbook with a HDD and SSD. In addition, we have also looked at the encryption performance using both types of drives.
With OpenSolaris maturing nicely, I would love to see some SSD and ZFS benchmarks. I think, among other things, the ability to set the recordsize of the ZFS system to match the native block size of a SSD will give ZFS a major advantage over other file systems using SSD.
Once btrfs is stablilised and more SSD benchmarks are run at phoronix, it'd be nice to see a CF/IDE thrown in the mix.
Most SSDs these days are being manufactured for SATA capability.
For someone who doesn't have SATA or even a PCI slot to add in SATA, this is a good way to add SSD in an inexpensive way.
Were the SSD benchmarks run with NOOP schedular? Simplify switching to that gave me a 25% improvement in read/write speeds on my Ritek 32gb.
For anyone interested, here's related benchmarks done on acer aspire one (it has samsungs pata ssd) with several filesystems.
- The OCZ SSD is known to be slow when running under AHCI
- Random Writes are slow on most MLC SSD (the Intel is the only one I know of where this is not the case)
Your tests don't whos off SDD'd force
I'm sorry but all your tests target transfer rates, as opposed to seek time. EVen the Sqlite insert benchmark; you should have at least benchmarked multi-table swipe selects, with a DB size > RAM.
Hard Disk Drives have excellent transfer rates, and it's almost never been a problem in itself; seek times is what you spend the most time waiting for.
Too many CPU-intensive tasks, not enough latency testing
Yeah I think the dealbreaker with SSD is latency. You can see it clearly on the sqlite benchmark: it does a lot random short writes, and the SSD totally tanks.
I'm not convinced by most MLC SSD's out there (except the intel one) until they at least equal hard drives on latency. Honestly, I'd trade a bit of top speed for less latency every day of the week.