fio can do it all, but it does take a little bit of learning- not just how the tool works but WHY and WHAT all those switches are for. How deep do you want to go down the rabbit hole? start by reading the man page and maybe some yt videos...
With plp in mind. Would the fact that - all servers have multiple PSUs and are backed up by a UPS stack, and we have supply detection (not sure of the official bame for this) that automatically fires up our onsite generator once it detects the UPS's are under load - mitigate some of the risk? I guess motherboard failure is the most likely cause of the PCI cards loosing power.
With plp in mind. Would the fact that - all servers have multiple PSUs and are backed up by a UPS stack, and we have supply detection (not sure of the official bame for this) that automatically fires up our onsite generator once it detects the UPS's are under load - mitigate some of the risk? I guess motherboard failure is the most likely cause of the PCI cards loosing power.
As far I know PLP are not so much about Power loss but Performance and durability. Consumer ssds needs to write data more often since data would he lost in case of a power failure leadibg to a higher write amplification, faster wearout and worse performance. PLP devices can buffer the writes resulting in a speedup and less writes. I might be wrong but if this is true your UPS and PSU won't help you much to get more out of your consumer SSDs. To have them is still a good idea though for an actual power outage.
PLP seems like such an innocuous feature, but you need to look at it in totality of the device. SSDs by their nature do not like a lot of small writes, so to alleviate this they put a dram cache layer in front to facilitate turning them into batch writes- much in the same way a raid controller might. the dram cache layer insulates (at least in part) from the effects of garbage collection and other internal housekeeping. For consumer devices, the cache layer is relatively small, and consequently the impact of a power loss is relatively small- but so is the cache impact on writes. The inclusion of a plp capacitor enables the deployment of a MUCH LARGER cache.
To see the consequences of this in practice, you can set up a simple test- write continuously to the drive for a period of 24 hours, and note when the write performance goes from gigs per second to 50MB/S every few hours.
That aside, enterprise drives firmware prioritizes endurance and not high peak spiky performance more typical of a non server use.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.