Subject: Re: [vserver] HASH FS
From: "Jun OKAJIMA" <okajima@digitalinfra.co.jp>
Date: Sat, 26 Jul 2008 04:08:03 +0900

Thank you for your help.
Yes, basically, your explanation is right.
Sorry for my bad English.

             --- Okajima, Jun. Tokyo, Japan.


2008/7/25 Michael S. Zick <mszick@morethan.org>:
> On Fri July 25 2008 00:50, Jun OKAJIMA wrote:
>> I am not sure I answered all the questions,
>> anyway, I updated the web site.
>> http://www.digitalinfra.co.jp/20080720/hashfs.20080723.html
>>
>
> Looking through that link, perhaps I can help with some talk;
>
> The filesystem is indexed by a content identification of the target (blob).
>
> At the index level, it is a flat file system (at least the index is).
>
> Using a secure hash function to generate the index from the
> content ensures there is only a single copy of the content present.
>
> One example appeared in the Advanced Bash Scripting Guide (by myself)
> prior to the development of GIT for indexing file changes.
>
> Another example, using the contents of the ASUS EeePC source release
> as the blobs, can be found here:
> http://sourceforge.net/project/showfiles.php?group_id=211455
> Those are just the content identification base indexes; in both
> single text line records or as OpenOffice Spreadsheet files.
> (again, by myself).
>
> Access to the target by name (a named file system) is secondary;
> the content ID _is the name_ to interface with people and programs
> using a pathname/filename structure is an indirect operation - it
> actually translates pathname/filename to current content ID index.
>
> Changes to the target (blob) content; which of course changes the
> content ID, causes the pathname/filename access to index a list of
> content IDs; the current one at the head, prior versions (file revisions)
> following in the list.
>
> I have never implemented a complete file system from the above,
> but Jun OKAJIMA is making an attempt at it.
>
> The posted sourceforge files shows, as single line records, the
> relationship between any content id and the path/file name of the blob.
>
> Mike
>> Probably, the page does not tell all of
>> your questions, so ask me freely if
>> you have any problem.
>>
>
>> Especially, I dont know much of vserver like you,
>> so, using vserver with HASH FS is
>> what I need your opinions.
>>
>> And, in the long run,
>> you would be able to enter VE like this.
>> # vserver enter http://vserver.debian.org/etch
>> Your opinion is?
>>
>>             --- Okajima, Jun. Tokyo, Japan.
>>            http://www.machboot.com/
>>            http://www.digitalinfra.co.jp/
>>
>>
>>
>>
>> 2008/7/25 John Alberts <john.m.alberts@gmail.com>:
>> > Hi.  Thanks for the updated documentation.  I've read through the doc
>> > and I'm still pretty confused on how to use it in relation to vserver.
>> >
>> > For instance,
>> >
>> > In the documentation, you have the following:
>> > Make an index and blobs.
>> > # mount -o loop,ro -t iso9660 sample.iso /mnt
>> > $ cp -a /mnt aaa
>> > # chown okajima:okajima /blob
>> > $ find aaa | ./hashify.pl
>> >
>> > This whole section is confusing to me.  What are we doing here and why?
>> > What is the sample.iso in this case and why am I mounting it to /mnt?
>> > As this applies to vservers, vservers are usually stored under
>> > /vservers.  Would we want the entire /vservers be one hashfs
>> > partition, or does each guest vserver under /vserver need to be a
>> > seperate hashfs partition?
>> >
>> > Regards,
>> > John
>> >
>> >
>> > with cp -a /mnt/aaa command, what directory are we supposed to be in?
>> > What user are we supposed to use for the chown command since obviously
>> > my machine doesn't have a okajima account on it?  Can it just be any
>> > regular user or should it be root?
>> >
>> > find aaa | ./hashify.pl command, I guess here we are passing each file
>> > that we just copied to the directory called aaa to the hashify.pl
>> > script?
>> >
>> >
>> >
>> >
>> > On Wed, Jul 23, 2008 at 10:31 PM, Jun OKAJIMA
>> > <okajima@digitalinfra.co.jp> wrote:
>> >> 2008/7/22 Jun OKAJIMA <okajima@digitalinfra.co.jp>:
>> >>> For ML issue, the answer is "not for now".
>> >>> Of course, I know I have to make it before very long.
>> >>>
>> >>> If you have to contact me just now,
>> >>> using vserver ML or OpenVZ ML.
>> >>> I subscribe them all.
>> >>>
>> >>> For implementation docs, I am writing now.
>> >>> just wait a bit.
>> >>>
>> >>
>> >> Done. See this.
>> >> http://www.digitalinfra.co.jp/20080720/hashfs.20080721.html
>> >>
>> >>            --- Okajima, Jun. Tokyo, Japan.
>> >>           http://www.machboot.com/
>> >>           http://www.digitalinfra.co.jp/
>> >>
>> >>
>> >>>
>> >>
>> >
>> >
>> >
>> > --
>> > John Alberts
>> >
>>
>>
>