Re: Disk: Percent full a performance factor?

rob@xxxxxxxxx wrote on Mon, 20 Oct 2014 18:29:10 GMT:

Now that we're measuring disk in TB and MB is nought but a
rounding error does 80% still hold true? If so, why?

The reason for the recommendation is based on HDD performance
characteristics. Rotational delay and seek distance add
latencies when trying to move from track to track on the
platters.

It can be argued this does not hold true for SSDs due to no
latency to move from "track to track" as there are no tracks and
nothing spins! But 100% SSD configurations are still the rarity
rather than the norm so changing the guideline likely won't
happen until 100% HDD configurations become a rarity.

I believe you need to consider your workload's unique
characteristics for creating new stuff, enlarging existing
stuff, and deleting stuff regardless of whether you are 100%
HDD, 100% SSD, or some combination of the two. Trying to find
extents or logical block addresses when creating and extending
can create performance issues when over 80-85% full. Then you
add the latency from HDDs and you see where the guidance comes
from.

As an aside .... I have one customer who has an LPAR which can
run at 94-96% full without any performance penalty. We
benchmarked with a range of %full from 70 to 95% and 95% ran as
effectively as 70%. Another LPAR for the same customer suffers
horribly as soon as the 83.5% mark is crossed. Again, tested
and validated.

File size for critical files which may require a 2nd copy on
disk for some reason is also important to consider. What if you
decide to increase %full to 90% but a copy of your largest file
won't fit in the remaining GBs and it is critical to make a copy
for some reason? There are many ways to slice and dice this
"what if" question. What makes sense for your business may be
very different than what makes sense for someone else.