Re: Questions about placing TASS images into on-line database

[Tass Mailing List <tass@mail.alembic.net>]

Date: Tue, 15 Jan 2008 21:11:51 -0700
From: Robert <robert@logicalchaos.org>
To: Tass Mailing List <tass@mail.alembic.net>
Cc: richmond@stupendous.cis.rit.edu
Subject: Re: Questions about placing TASS images into on-line database
Resent-Date: Wed, 16 Jan 2008 06:55:17 -0700
Resent-From: Robert <robert_creager@logicalchaos.org>
Resent-To: Tass Mailing List <tass@mail.alembic.net>

Tass Mailing List wrote:
>
> Date: Tue, 15 Jan 2008 09:43:39 -0500
> From: Michael Richmond <richmond@stupendous.cis.rit.edu>
> To: tass@mail.alembic.net
> Cc: mwrsps@rit.edu
> Subject: Questions about placing TASS images into on-line database
>
>
>   I can handle the correction of the raw images, and probably
> the creation of co-added images (if that becomes reasonable).
> However, I don't have much experience with such a large data
> volume.  Can others on the mailing list help?
This is a small data set these days.
>
>   What I'd like to know is:
>
>      - what sort of hardware would you recommend for the job?
Any old off the shelf computer.
>
>      - should I simple purchase a unit which can store all the
>             information (like an NAS), or try to build one
>             by purchasing separately drives, an enclosure,
>             and perhaps a controller card?
I'd build one (I did).
>
>      - hardware or software RAID?  I'm leaning towards software,
>             but seek the voices of experience
Software Raid, with XFS on LVM on top of the raid.  This will allow for
disk failures and for growing your storage easily (with new drives).
>
>      - any suggestions for the software to connect a database
>             to a web browser?  Note that this database
>             will be read-only
NFS mount will do fine, if you don't just put the disks on the web
server machine.
>
>   Suggestions?  Comments?  Guesses at an overall cost?  I welcome
> all input.  If we start to go into really gory details, it might
> make sense to take the discussion off the E-mail list and conduct
> it via private messages, but let's see if that becomes necessary.
Presuming you have a computer with 4 empty disk slots (3 1/2) and one
controller card slot, it would cost under $1500 for 4 1Tb SATA disks
with a dedicated controller card.

For my gory details.  I put a system together about a year ago for doing
mythtv, and TASS data storage (if I ever get ROB going again).  I put in
6 disks.  2 320Gb disks Raid 1 for the boot disk/root partitions, and 4
400Gb disks Raid 5 for the data storage (1.2Tb).  I can record 2 shows
and display another, while running Boinc with no problems :-)  Video
data demands are about 2Gb/hour.  The machine is a dual core AMD 64
5000+ with 2Gb memory running Ubuntu 7.10.

I've included two bonnie++ runs on the raid1 and raid5 subsystems with
default parameters below.  I can run more tests if you like (or just
multiple runs to average out any glitches), and through a NFS mount on a
Gb link to another server if you're interested.

Raid 1:
Version  1.03       ------Sequential Output------ --Sequential Input-
--Random-
                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
--Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
/sec %CP
mystic           4G 40434  60 43179  10 20590   3 43186  60 46044   5
274.9   1
                     ------Sequential Create------ --------Random
Create--------
                     -Create-- --Read--- -Delete-- -Create-- --Read---
-Delete--
               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
/sec %CP
                  16  1820   3 +++++ +++  4229  18  3115  12 +++++ +++
1245   6
mystic,4G,40434,60,43179,10,20590,3,43186,60,46044,5,274.9,1,16,1820,3,+++++,+++,4229,18,3115,12,+++++,+++,1245,6

Basically, 40Mb/sec read/write, 20Mb/sec rewrite.

Raid 5:
Version  1.03       ------Sequential Output------ --Sequential Input-
--Random-
                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
--Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
/sec %CP
mystic           4G 50915  76 72346  15 37265  15 52334  89 172670  64
321.9   1
                     ------Sequential Create------ --------Random
Create--------
                     -Create-- --Read--- -Delete-- -Create-- --Read---
-Delete--
               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
/sec %CP
                  16  1554   5 +++++ +++  1784   7  1620   4 +++++ +++
440   1
mystic,4G,50915,76,72346,15,37265,15,52334,89,172670,64,321.9,1,16,1554,5,+++++,+++,1784,7,1620,4,+++++,+++,440,1

Even faster - 37Mb/s to 72Mb/s, until the sequential block input which
hits 172Mb/s?  Dang, that surprises me a little, but that's what you
would be looking at reading an entire image off the disk in one hit.

Cheers,
Rob