Boost logo

Boost :

Subject: Re: [boost] This AFIO review (was: Re: [afio] AFIO review postponed till Monday)
From: Niall Douglas (s_sourceforge_at_[hidden])
Date: 2015-08-23 11:52:12


On 23 Aug 2015 at 11:16, Glen Fernandes wrote:

> On Sun, Aug 23, 2015, Niall Douglas wrote:
> > I couldn't find a Table 1.4.
> >
> > Did you mean Table 1.5, the one comparing NoSQL databases with the
> > naïve iostreams and AFIO key-value store and where AFIO comes off
> > looking particularly badly?
>
> "Table 1.4. This solution will perform reasonably well under these conditions"
> http://boostgsoc13.github.io/boost.afio/doc/html/afio/quickstart/workshop/naive_afio.html
>
> That "Maximum performance isn't important to you" was crossed implying
> that this solution satisfied the desire for performance. Unless you
> meant "is" instead of "isn't", but then the benchmarks don't reflect
> that.

The idea behind the workshop tutorial is we start with a really
simple design which is one file per key-value, and move onwards from
there with designs of increasing complexity and sophistication all
implementing a key-value store with various improving characteristics
like whether the store will survive power loss. The table at the end
of each of these lists the characteristics of that particular step
with a green tick (good) or a red cross (bad).

One file per key-value is always going to be slow under any
circumstances. Hence the "Maximum performance isn't important to you"
red cross.

The final design should have pretty close to constant time
performance for lookup and linear complexity non-concurrent modify,
but the implementation is very considerably more complex than a one
file per key-value design.

Niall

-- 
ned Productions Limited Consulting
http://www.nedproductions.biz/ 
http://ie.linkedin.com/in/nialldouglas/



Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk