HN Explorer

5 comments

[ 3.3 ms ] story [ 23.3 ms ] thread
jpadvo
"The idea is to focus on simplicity and performance. Not availability or fault tolerance (at all)."

This sounds absolutely perfect for a restricted number of use cases, and really tantalizing but unworkable for the rest.

scottmp10
I agree here. I understand the trade offs being made but I just don't see this being useful for too many people with zero fault tolerance and the single point of failure meta data master. It probably gets some pretty good performance improvements from not worrying about that though.
mark242
If you are designing a filesystem that does not handle fault tolerance, you basically have N single points of failure. It's essentially the difference between RAID 0 and RAID 5.
jkempf
"Distributed single point of failure" is the term we use where I work.
eropple
I have a hard time calling something without fault tolerance "done right." (S'why you don't exactly see RAID 0 all over the place.)