In other words, your write amplification would increase from 30-50 for
normal LCS to 200-300 when fan-out changes to 100.
On 18/09/2018 14.54, Marcus Eriksson wrote:
problem would be that for every file you flush, you would recompact
all of L1 - files are flushed to L0, then compacted together with all
overlapping files in L1.
On Tue, Sep 18, 2018 at 4:53 AM 健 戴 <daij...@me.com.invalid> wrote:
Hi,
I have one table having 2T data saved in c* each node.
And if using|LCS|, the data will have 5 level:
* L1: 160M * 10 = 1.6G
* L2: 1.6G * 10 = 16G
* L3: 16G * 10 = 160G
* L4: 160G * 10 = 1.6T
* L5: 1.6T * 10 = 16T
When I looking into the source code, I found an option: |fanout_size|.
The default value is 10. What about change this value to 100? Then
the level will reduce to 3:
* L1: 160M * 100 = 16G
* L2: 16G * 100 = 1.6T
* L3: 1.6T * 100 = 160T
Or even could I set this to 10000? And all files are in a same
level. Should it be better then?
What is the cons of the bigger value of this option?
Thanks for your help.
Jian
