Re: How about SAN with ARM?
Hi again and thanks for writing.
I got pretty busy and procrastinated to reply.
Please forgive me.
On 10/16/2012 09:52 AM, Luke Kenneth Casson Leighton wrote:
On Tue, Oct 16, 2012 at 1:08 AM, Muun Dahweed wrote:
I am trying now to get the web scripts working that will allow connecting
the storage as a typical NDAS from my sheevaplug, but it looks like there is
no big market now. So I am asking what users here think of this.
Shall I continue? Do you think the growing power of ARM boards will lead to
simply building in HD and such? Or would the future lead to scale out type
connectivity, using several block devices on LAN as a storage for lots of
ARMs doing background storage and service.
for this to take off in any big way, i think someone needs to be
"sold" the idea where it would make a significant impact - a dramatic
improvement in profitability, to give them that competitive edge.
I was considering that the lack of posts about it as a system kind of
hinted that it is not practical. However, I am just trying to see things
in historical context. As in, 30 years ago, the computer filled the
room, but now is in your hand ....
Well, today, the big computers and huge storage seem to use SAN and
Block Level pretty heavily. So perhaps in the near future it will be as
useful in the home.
i figured that the much simpler way would be web service farms of
LAMP/LAPP systems [with oracle's reputation going to shit i prefer to
promote postgresql], where the ARM systems cover the LA.P and the
standard big-iron systems cover the SQL server part. this would be
highly effective in areas where high latency is not a real big issue.
I did not even think of databases. I was just thinking of archives and
redundant backups, using something like the pooling software mentioned
later.
the nice thing about this is that web server farms already recognise
and deploy the concept of round-robin DNS or HTTP proxy redirection in
order to farm out the queries to individual web front-ends.
Over my head. Sorry.
where was that article last week which explained the horrendous
amount of power being wasted on x86 systems just *idling*? those
systems have to be spec'd to peak throughput. embarrassing or what.
just so we can get at "da clowuhhd". a pack of 5 ultra-low-power ARM
systems replacing one beefy x86 system would have both performance
*and* power usage wayyy better.
Lot's of power saved by Arm. Ok. So one reason to keep at my project.
Big data from the Little server! it sounds like a sales pitch from heaven!
i'm intrigued about the NDAS idea though. i looked up NDAS, there's
a company called ximeta - they apparently released GPL linux kernel
drivers for their proprietary protocol, but code.ximeta.com has been
taken offline, since.
The technology was bought last year. And the new owner opens the
sources. Now it moves even to git hub and aims for submission to Linux
driver staging. However, as far as I know, the advent of the "Clouwd"
also put a kibosh on the home user personal need for storage, and the
hardware in the home seems a little less popular for now.
It is hard to know if it will change. This is why I made this inquiry.
what did you have in mind, and what, overall, is the problem that
you're looking to solve? are you envisioning little ARM computers
with SATA drives (rack-mounted, 1U, Qty 16 per rack), several of these
per cabinet, then being turned into a "Global File System", like this:
http://rhombus-tech.net/community_ideas/cluster_server/
l.
There is no NDAS rack yet. Besides, I see another block device provider
or two that have some fine devices like that. I did try one of the
software disks in a test, so I know it does work from the arm computer
too. I hoped to meet a user with a few plug-puters having the two
network ports, and some experience with GFS to maybe put it to the test.
This is the only tests I have seen so far and they are rather old.
http://ndas4linux.iocellnetworks.com/kermit/index.cgi/wiki/HowToGFS
http://ndas4linux.iocellnetworks.com/kermit/index.cgi/wiki/HowToOCFS2
As for my specific problem. I could not find an cloud system readily
working with NDAS from my debian Sheevaplug. When I did "apt-cache
search cloud" there was not much to use. A pogo plug was fine, but hacky
to make the netdisks usable, and the goflex was fine too, but also
hacky. They required using the optware system.
I wanted something much easier and more standard. Now I am tinkering
with a cloud package that was hidden in the NDAS driver codes. It is
getting almost usable as PHP scripts to manage the NDAS drives from the
web, and soon should be able to get all the other functions up and
running. Format drive, partition and such. Then see what goes from there.
I guess one way a GFS or OCFS2 implementation would be useful is for
some low budget classroom runs a bunch of $150 Arm based desktops all
connected to a 100 dollar NDAS as the central storage. There might be
other applications though.
I started on this when I saw one other NAS project running Greyhole,
from an ARM computer which seemed like a good match for many NetDISKs on
the lan, all connected to a single server and used as as pool with
redundancy via SAMBA. However, it looks like that Arm based side was
dropped. They are still going on Ubuntu for 86's though.
That's a summary of my status. Thanks again for you encouraging words. I
will keep going, even though if only a slim chance.
Reply to: