From daniel at phunq.net  Mon Sep  1 03:08:09 2014
From: daniel at phunq.net (Daniel Phillips)
Date: Mon, 01 Sep 2014 03:08:09 -0700
Subject: MVRB-Tree vs PHTree
In-Reply-To: <87bnr78ek3.fsf@devron.myhome.or.jp>
References: <CALU6X8m_H85mRE88msLveq7PhY94UMJ+Wj-0qFS_ZFHH08ebgg@mail.gmail.com>
 <87bnr78ek3.fsf@devron.myhome.or.jp>
Message-ID: <13c2f44b-3dba-49b9-9914-d75ede282145@phunq.net>

On Tuesday, August 26, 2014 8:18:36 AM PDT, OGAWA Hirofumi wrote:
> Elifarley Callado Coelho Cruz <elifarley at gmail.com> writes:
>
>> Hi,
>> 
>> I wonder if you guys see any possible benefit of using MVRB-Tree
>> instead of PHTree as a data structure for Tux3. It's used in the
>> OrientDB NoSQL database.
>> 
>> From https://github.com/orientechnologies/orientdb:
>  ...
>
> Hm, I'm still not sure about the detail of MVRB-Tree though. MVRB-Tree
> seems to be deprecated by OrientDB since v1.6.
>
> http://www.orientechnologies.com/docs/last/orientdb.wiki/Indexes.html

Hi Eli,

I agree with Hirofumi. Not having durability would immediately disqualify 
MVRB
as an indexing scheme for a file system. However, you will be happy to hear
that phtree is already obsoleted in favor of shardmap, a much better 
indexing
scheme that appears to scale comfortably into the billions of files per
directory, and unlike MVRB, is durable (durability is actually the hard 
part).
See the prototype in our Git tree. Hopefully I will present a talk on 
Shardmap
in the not too distant future, and also turn the prototype into production 
code.

Regards,

Daniel