Gluster users - Nov 2010 - Concepts of Gluster: Block- vs File-based | How does cluster internally work?

Hello Experts,

I would be thankfull for an explanation, what concepts are behind gluster? I
would like to run a database over gluster. As I have read in wikipedia,
running Databases on top of "Block-based Storage Concepts" (like
iSCSI)
works quite well. However it seems that Gluster is conceptually nearer to
the NAS Concept with distributing whole files?

However never the less Cluster needs internally also some concepts to
provide random file access...

Some of my Questions:

1.) When requesting a file, is the whole file retrieved or only a portion of
the file?
2.) Are random File accesses supported?
3.) How does cluster internally Workd when the Database (liek MySQL) create
a large file and then accesses this file? Does Cluster shomehow operate on a
"Block Level"? How will an index-scan be translated in reagards how
gluster
will access the data on the remote system?

I am very thankfull for any information (also links...) that help me to
understand how gluster internally handles "file accesses" .

(I have reas the "Architecture Document" of Gluster (dont know the
exact
name, was hosted on some ftp server of gluster) but this document only
contains marketing language and does not go indepth...)

Thank you very much!!
Jens

Hi Jens,

Read Inline..

> Some of my Questions:
>
> 1.) When requesting a file, is the whole file retrieved or only a portion
> of
> the file?
>
Gluster has 1-1 mapping between client side posix calls to server side posix
calls. ie,
when a open is called on client mount point, it will be translated to a open
on server
side (where the file actually resides). Answer for your question will be you
will get the
portion of file you requested for, and not whole file.

> 2.) Are random File accesses supported?
>
Yes. But, considering scenario in case of Databases, which is, random file
accesses with
query(read) for very small blocks, network delay and context switch delays
involved will
make the performance suffer.

> 3.) How does cluster internally Workd when the Database (liek MySQL) create
> a large file and then accesses this file? Does Cluster shomehow operate on
> a
> "Block Level"? How will an index-scan be translated in reagards
how gluster
> will access the data on the remote system?
>
>
As I told in case of 1, Gluster will do 1-1 operation which is done @ the
mount point
over the server, hence the main issue again will be latency of the network
in all these
operations.


> I am very thankfull for any information (also links...) that help me to
> understand how gluster internally handles "file accesses" .
>
>
Lets be clear on what 'file accesses' means. I am taking it as
'read'
(considering its a
database and there will be lot of queries), but there are multiple system
calls which can
mean the same with respect to filesystem, like, access(), lookup(), stat(),
open()...


Regards,
Amar