Rocks Cluster Tutorial Part 1 The Install

[Airbozo]

==NOTE== Please keep in mind that this is a work in progress, so if you see something that does not make sense, ASK QUESTIONS! Also if you see anything that is confusing and could be better written, please let me know. And if I have made some mistakes also let me know. I am still learning this stuff too. ==NOTE==


This is the start of the Cluster Tutorial using Rocks Clusters Software;
http://www.rocksclusters.org/

This tutorial is based on the software from Rocksclusters.org. I will be refering to some of the documents on their site, so if you want to follow along, please point your browser to; http://www.rocksclusters.org/rocks-documentation/4.2.1/index.html. You can also download a pdf version of this document to view off line.

Getting started.
The basic items needed for a simple cluster are a frontend node (the system that the rest of the world will see), a compute node (1 or more), a private network interconnect, and an external network interconnect, Software.

HARDWARE;

Frontend (1);
motherboard
processor (386 minimum)
memory (minimum 512mb)
Hard drive (20gb min)
CD drive (dvd for the larger disks)
2 network interfaces (one for the outside world and one for the cluster network)
Power supply (to power all those jiggly bits)
Video card (not really necessary since _most_ systems can use the serial port for a console)

Compute Node (at least 1);
motherboard
processor (386 minimum)
memory (minimum 512mb)
Hard drive (20gb min)
CD drive (dvd for the larger disks)
1 network interface
Power supply (to power all those jiggly bits)
Video card (not really necessary since _most_ systems can use the serial port for a console)

Network;
Switch for connecting cluster together. Can be eliminated for a dual node cluster and a crossover cable.
Cables to connect all systems.

Normally the frontend node is connected to the outside network since this is the unit that will be "seen" by the world and is responsible for communication to the various nodes and task handling, etc.

While it is not necessary for all nodes (including the frontend) to be the same hardware it makes things easier if they are close. For example, try not to mix different processor types (amd's, pentiums, xeon, Itaniums, etc). It is also alot easier if the network cards in the different nodes are the same or at least from the same manufacturer.

Here is the example of the network layout;


http://www.lotechdesigns.com/host/images/4423Cluster network.jpeg


SOFTWARE;

For the first tutorial I will just be installing the 2 systems as a web server cluster, so the following software will be needed;

Kernel/Boot Roll CD
Base Roll CD (also called the Core Roll)
HPC Roll CD
Service Pack Roll CD
Web Server Roll CD
OS Roll CD Disk 1
OS Roll CD Disk 2

The "Rolls" for this project can be found here;
http://www.rocksclusters.org/wordpress/?page_id=3

If you want to follow along with this tutorial, download and burn the ISO's to disk, then return here.
While the ISO's are downloading and burning, it is a great time to browse through the documentation, here;
http://www.rocksclusters.org/rocks-documentation/4.2.1/index.html

Since most of you out there already know how to hookup a computer I will not be covering that. Make your frontend boot ready and set the first boot device to the CDROM. Gather your install "Rolls" and get ready.

Up Next: Installing the Frontend!


EDIT: Got swamped at work so am still working on the second install on this...(but I know you guys still need time to gather items... ;) )

[Luke122]

Thanks again for this tut, I'm gathering the needed stuff right now. I've got 4 p3-866's w/512mb and 40gb hdds ready to go for this project. All identical hardware too. :D

Getting the software now, and lookin forward to your next post.

-Luke

[Airbozo]

Once I clear my workbench I am prepared to continue.

My Boss was on vacation for the last month and I had to take up the slack. I just cleared the last big item from the bench (a failed Cintiq display) and should be able to stage my cluster nodes to continue this tutorial.

I am also re-reading the "official" documentation to make sure I am not just doing a copy and paste. I have a fairly good working relationship with these folks and do not want to piss them off. They have helped me a bunch in the past so I am treading lightly. A few people from the development team are watching this thread too...

[Luke122]

Well thank them for me also, for allowing you to do this tutorial for all of us here at TBCS.

Also, I've decided to pass on the p3s, as I have 4 -P4 1.6's here to be used for "something" instead. :D Should be a potent little cluster.

Regarding the Cintiq, do you repair LCD panels also? What type of training experience would you recommend to get into that sort of thing? Our local tv repair shop has been yanking me for quite awhile to repair my 27" hdtv, and I'd love to be able to stick it to them and fix it myself.

[Airbozo]

I don't really do any LCD repair, nor any TV repair anymore. It seems the more experienced I get, the less component level repair I do. I actually started out fixing TV's, stereo's and musical equipment (Amp's, mixers, effects peddles, etc), for my adopted Dad's shop.

Basic Electronics is a good place to start. If you like working with the board level stuff, learn how to solder. Then learn SMT techniques.

One thing with the Wacom Cintiq, is that Wacom tech support is kind of anal about RMA's. They basically refuse to work with you until you send a picture of the problem. Vertical blue line on right side of display, is not good enough, they need visuals...


EDIT: Oh and the guys at UCSD didn't really "allow" me to do this tutorial, they were kind of baffled why I didn't just use theirs... They did ask me not to cut and paste though and I do respect that.

[Commando]

Thanks for this tutorial. It's always great to see some solid content being shared.

[chaksq]

Would it be a bad idea to have the frontend with worse specs than the nodes?

I have a couple of gateway lying around. 2 Gateway 6400 Servers and a budget Gateway Essential. I might try this just for kicks. My setup would either be with the Essential as the Frontend and the 6400s as the nodes or only use the 2 6400s with one frontend and one the node.

None of this is much use to me now. I just figure this could be an interesting project. My best bet would probablt be to use the two 6400s and set up for redundancey in case one fails. Nothing else is much use to me. Other than one having 2 extra HDs they are set up identicle. Maybe I could use it as a web server or something.

You said to ask questions so there you go. What is your imput?

F.Y.I. both the 6400s are running fresh installs of Ubunto Linux and the Essential needs to be reformated.

[Airbozo]

Would it be a bad idea to have the frontend with worse specs than the nodes?

I have a couple of gateway lying around. 2 Gateway 6400 Servers and a budget Gateway Essential. I might try this just for kicks. My setup would either be with the Essential as the Frontend and the 6400s as the nodes or only use the 2 6400s with one frontend and one the node.

None of this is much use to me now. I just figure this could be an interesting project. My best bet would probablt be to use the two 6400s and set up for redundancey in case one fails. Nothing else is much use to me. Other than one having 2 extra HDs they are set up identicle. Maybe I could use it as a web server or something.

You said to ask questions so there you go. What is your imput?

F.Y.I. both the 6400s are running fresh installs of Ubunto Linux and the Essential needs to be reformated.

The front end is usually the same or better than the slave nodes. Mainly because the front end is responsible for so much more than just pumping out visuals. The visual nodes basically are just graphics pipes doing little processing, besides local i/o, networking, and some transformations. The front end actually sends all the data to the individual nodes and makes sure they are on task (and on the same task), as well as determining _what_ should be sent to each node (it calculates the view port for the rest of the nodes).

That said, unless you are pumping out high quality video with a moving view port and multiple viewing angles, it really will not matter.

I am still working on this and am currently running through a trial install. I want to make sure that everything is working right before posting. I did a major noob bungle... I got all my hardware setup, found some similar video cards, cleared my workbench, got the camera setup, only to find out I downloaded the amd64 version of the software, and my proc's are NOT amd64/emt64 compatible... Rather than download yet another version, I pulled some other test systems I have for this project. So it turns out that my initial web server cluster will be on two identical systems, but when I do the visual cluster, I will use a different front end.

Here is where I am going next week;

Rocks-A-Palooza III (http://www.rocksclusters.org/wordpress/?p=50)

I have asked one of my friends that works at ATI (at the AMD headquarters), for a tour of the facilities when I am there.

[nil8]

Have fun touring AMD! Lucky.

The only thing holding me back from getting this going with some old machines is lack of KVM cables(I need 2 more) and this tut. I've always wanted to try clusters and this looks like the place to start.

Thanks for this Airbozo. We really appreciate it.

[Airbozo]

nil8, Don't let the lack of cables hold you back. Once the frontend is installed and setup you can disconnect the monitor and KB/mouse and connect them to the first node (while it is running). Once the second node has booted and finished install you can move the cables to the next system (and so on...). Once you are done with the software install you can reconnect to the front end node and that is all you will need, since it will be _one_ cluster. If need be you can then use a serial console to access the systems.

[nil8]

I'm placing my order this weekend and working on buying a couple more bitch boxes to play with. Mine are starting to show their age. At least 1 for the front end and my next attempt at immersion cooling eventually.

As for the KVM cables, I need them anyway. I have a 4 port kvm and other uses for it at home.

512mb ram min for everything is a but much isn't it?

It's good to know that everything can be accessed via serial.
How difficult will this really be? What I mean is roughly how many hours is it going to take to setup and get operating?

[chaksq]

Well using the systems I had planned on is not going to happen. I ended up upgrading one and using the other for parts.

However I had an idea. Would it be possible to create a cluster using thin clients? They are fairly cheap used and very low power. Too qualities that appeal to me greatly.

[nil8]

Short answer:98% no

Long answer:
Thin clients are shells of machines that can locally load specific protocols designed to remote into servers, such as Citrix ICA or MS RDP.
Beyond this and a couple of local features, there isn't anything to them.
Their sole purpose in life is to bring a server session to your eyes and hands. They don't process (much) data and have extremely small amounts of ram. All that's needed is the stripped out OS. The common flavors consist of linux or Windows CE.

Your best bet if you actually want lower power consumption and keeping a decent price range is 2nd gen sff machines. Around the 1.5ghz mark will serve you well and run you less than 50 a box. Thin clients can be more expensive than this, even 2nd hand because they're not common anymore.

[Airbozo]

If it can boot linux, you can cluster it. You can even create diskless nodes (but I will not be getting into that). During the install all you have to do is replace the OS roll disks with your favorite linux distro (most all will work). I was talking to a friend who said they even made some systems that have linux on a chip into clusters. They had to go through the trouble of flashing new roms, but once they figured everything out it works fine. I think they are going to productize it soon.

As far as the time to install, I will post some approximate times for each stage and each type of node (compute/visual, etc). It took me about an hour to install the frontend and less than that for each node (since you don't have to feed the slave nodes any cd's).

[Drum Thumper]

Compute Node (at least 1);
motherboard
processor (386 minimum)
memory (minimum 512mb)
Hard drive (20gb min)
CD drive (dvd for the larger disks)
1 network interface
Power supply (to power all those jiggly bits)
Video card (not really necessary since _most_ systems can use the serial port for a console)



A 386 series processor who's mobo can handle 512 megs of ram? And a BIOS that old that will handle a 20 gig hard drive?

I'm gonna have to google that, but I don't think that is feasibly possible.

+rep for the thread--gives me an excuse to get more computers!

[Airbozo]

...Almost done... Damn work... Keeping me from finishing this tutorial.

I am tempted to take these systems home and finish there. (hehe the SO would kill me if I brought another computer in my house even if it is temporary. Yeah right, she's heard that before...)

[nil8]

Patience is a virtue of IT. It will be, in time. Work on your list of stuff to do bozo. We're not going anywhere.

[Airbozo]

Amazing! I actually have time to work on this! I am starting the install now and am trying to figure out how to take screen shots, otherwise I will just photograph the monitor.


TT

[Crazy Buddhist]

Would it be a bad idea to have the frontend with worse specs than the nodes?

...

F.Y.I. both the 6400s are running fresh installs of Ubunto Linux and the Essential needs to be reformated.

Did you give up on this? It's really easy to get clustering working within Linux, from my post in this thread,
(http://www.thebestcasescenario.com/forum/showpost.php?p=188881&postcount=7) if you follow the link to the microwulf cluster pages, the guy explains a slightly complex way of getting ubuntu to cluster.

I'd have one of your high spec machines as the principal machine and any other machine you can lay your hands on as nodes. Also check out clusterknoppix (http://clusterknoppix.sw.be/) - a linux clustering distro that can boot off the CD drive and have a cluster up and running quicker than I typed this post.

CrazyB

ps Mr AirBozo .... what happened to your Cluster? Life get in the way?

[nnigam]

[QUOTE=Airbozo;84807]==NOTE== Please keep in mind that this is a work in progress, so if you see something that does not make sense, ASK QUESTIONS! Also if you see anything that is confusing and could be better written, please let me know. And if I have made some mistakes also let me know. I am still learning this stuff too. ==NOTE==


This is the start of the Cluster Tutorial using Rocks Clusters Software;
http://www.rocksclusters.org/

This tutorial is based on the software from Rocksclusters.org. I will be refering to some of the documents on their site, so if you want to follow along, please point your browser to; http://www.rocksclusters.org/rocks-documentation/4.2.1/index.html. You can also download a pdf version of this document to view off line.

Dear AirBozo
I do not see any dates for these postings, so do not know how old this is. However, I do not see anything beyond the hardware/software requirements. I know that all of us are busy with our daily tasks to pay our bills, and you may not have had time to work on this for some time. Please let me know. I plan on installing a rocks cluster this week, and can provide help with your writeup if needed.

I do appreciate the link to the docmentation which is helpful, but I could not find a pdf version of this, and as I saw this as I was getting ready to board a flight, I could not read it on flight. However, will pick this up when I land and as I have time

Thanks
Neeraj

[nnigam]

Would it be possible to have a 32bit front end with 64bit compute machines?

[Airbozo]

Would it be possible to have a 32bit front end with 64bit compute machines?

Not sure about that since the compute machines get the code from the front end when installing. You may want to double check since I know there are many ways to setup the nodes. I would _assume_ that if the tools used to control the compute nodes were similar, they may work. Check the ROCKS site for more info.

[AlexGg]

Dear Airbozo, I have built latest version 6.1 of Rocks Cluster 32 bit with two just nodes, but as far as I am still new in this, I got a question - how to build my own website on the cluster.
Could you please or someone help me out with this?

All I need to know: where to put website files and activate to be viewed from LAN I have this cluster build in? Could it be visible from WAN/Internet?
Is there any GUI to create new web-server? Sorry if this question is dummy...