RAID-5 Hardware and Software Methods Compared

Discussion

Drawing Conclusions

The results of these tests are based on the hardware and software configurations I have described. How they apply to a different environment is unpredictable. When I build an environment, I validate that it provides the level of performance I expect or can, at least, tolerate. Keep in mind that your environment’s performance may not conform to these results.

If you find any discrepancies in these tests, or have recommendations as to how to improve them, leave a comment with your thoughts.

Ubuntu RAID-5

Although Ubuntu hardware RAID scored better in general, Ubuntu mdadm did a respectable job trying to keep up and even outscored hardware for random reads, if only by a bit. LVM scored best — again, marginally — in the random read patterns, and slightly better than mdadm in sequential reads, but did considerably worse with write performance. If hardware RAID were not an option, mdadm is the better performer.

Windows software and hardware RAID

The legacy software RAID method for Windows did okay. Read performance compared to hardware RAID was excellent, and it did even better than the hardware on sequential reads. This is a mystery because Windows hardware RAID did considerably worse reading sequentially than Ubuntu with the same hardware configuration. Write performance is substantially better on hardware in all three tests, especially sequential writes, so Windows hardware RAID was the best performer for this OS.

And Windows software RAID was better than Windows Storage Spaces, as will become clear in the following section.

Windows Storage Spaces Performance

The performance was pretty bad. Windows Storage Spaces did the worst in every category except 4 KiB random writes, where it came in second to last and with very wide variance. 64 KiB random writes were very slow and sequential write performance was abysmal.

If poor performance were not enough, the additional complexity involved in configuring (and fixing1I recently experimented with parity spaces and the experience convinced me to go to hardware RAID for that application. Not only was the performance bad, but the virtual disk refused to repair itself after a drive replacement. The Windows Storage Spaces GUI is all but useless in such cases, and while the PowerShell interface sometimes offers a clue as to why your virtual disk or pool is unhealthy, at a certain depth the product becomes opaque and the only option left is to spend a few days copying your data somewhere else and rebuilding from scratch. A quick search online will link you to myriad Storage Spaces horror stories.) Windows Storage Spaces makes it the least attractive storage method in this lineup.

Leave a comment

Your email address will not be published. Required fields are marked *