Slashdot Mirror
Building Your First Cluster?
An anonymous reader asks: "I'm interested in building a DIY cluster using Linux and will be using conventional Linux software. However, the number of possible ways to do this is huge. Aside from Beowulf, there's Mosix, OpenMosix, Kerrighed, Score, OpenSSI and countless others. Therein lies the problem. There are so many ways of clustering, development seems to be in fits and starts, most won't work on recent Linux kernels and there's no obvious way to mix-and-match. What have other people used? How good are the solutions out there?"
Try them all. After all, you just KNOW your first one's going to be a clusterf*ck.
Seriously, if you're going to take that route, you really should be prepared to invest the time in test-driving several different solutions.
It kinda depends upon what you want to do. What are the requirements for your cluster? Have you considered going massively SMP instead? Or are you just after a cluster for the sake of having one?
Rocks is a great tool to build a cluster. It includes lots of monitoring tools and such, so you can see the status of nodes, etc. However, I'm not quite sure how large you're planning on going... May be overkill for a 4-noder. =)
Sounds like you're trying to solve a problem that you don't have.
Cluster need special software to take advantage of the disturbed computing. They are built with a specific task in mind. Or do you already have a need and just failed to tell us?
For me, I run my network with distcc (http://distcc.samba.org/) So all of my Gentoo boxes can compile using shared computing power. It cut a typical 33Min app down to less then 2 mins doing this. And works wonders for my slower laptop.
With distcc, all you need to do is have the same tool chains. (glibc, gcc, coreutils, etc) You can even specify how many threads per box you want running to fine tune your network.
On the other hand, if you just want to learn, then you should try them all. The all suit different needs.
hella X-Boxes...
Beowulf... cluster. Must... be...
BEOWULF!!!
You may want to look at this online book (free):
http://linuxclusters.com/compute_clusters.html
At least get to know various approaches at a high level before proceeding...
Two computers make a 1 dimensional "cube." Four, in a round-robin make a square. Six, properly connected, make a regular cube and so-on. Does anybody out there know if they still connect clusters this way and if not, why?
Good, inexpensive web hosting
beowulf: http://www.rocksclusters.org/wordpress/ or http://www.warewulf-cluster.org/cgi-bin/trac.cgi
i'd recommend looking at the ease of building diskless clusters w/ warewulf.
I have a stack of five origional Pentium boxes with 32mb of RAM and 2gb harddrives (except for one, with a larger drive for a software repository). Origionally built it to experiment with AFAPI based clustering, but since AFAPI is a reasonably non-invasive setup, it works well for trying other techniques too, everthing from simply running distcc on the nodes to speed up i586 software builds to briefly fiddling about with some of the other clustering options mentioned. Fiddling around with options on a real cluster (running cluster software on a single node really isn't a good impression) that could be reinstalled from scratch in a few hours, and the machines aren't worth enough to matter if it is physically damanged is a great way to learn.
I have 5 boxes of somewhat random configuration that I am looking to do the same thing as the OP. My goal for this cluster would be high-speed video processing (using tovid).
"The price good men pay for indifference to public affairs is to be ruled by evil men." ~Plato (427-347 BC)
1. I do beowulf, other clusters arent my thing. 2. I can handle C and C++, but I'm not a guru. 3. I can fumble my way through unix-linux but I get cranky with new versions (command/ flag changes in utilities). 4. I have 6 lazyish years as a unix sysadmin.
getting prepped,
You want to make sure that the boxes that are talking to eachother are very secure from the rest of the world. Most of the concepts on a cluster are about trashing the security of the machines in question. There are ways to make a secureish cluster, but a good firewall is a better way to go. let the firewall talk to your "head node", and preferably locks the "body nodes" from seeing the outside world. There are a ton of ways to get this done. on the cheap have all the body nodes have a non-existant gateway ie 192.168.0.1, set the firewall as 192.168.0.129 (forget dhcp) and let the head node point to the 192.168.0.129,, and have the firewall route services (ssh, ftp, telnet (ok not ftp or telnet)) to the head node.
getting started
1. Load all the boxes with the same OS. (the same way) (DONT select SELINUX or you will cry) 2. build a hosts file (names for machines) /etc/hosts
3. build a hosts.allow , hosts.equiv (still in /etc)
4. add in some entries into securetty for your rlogin rcp rsh..
5. youll probably have kerberos(weakly secured) rlogin rcp rsh... you want to rename those and replace them with the non secure versions, there are other ways, but this saves a bunch of hassle.
6. pop into /etc/pam.d and adjust the rlogin rcp rsh.. (this may not be needed in some cases).
7. add in a + + in the .rhosts file of each cluster user.
after you have pulled your hair out decyphering my glossed over instructions, you should be able to type: rsh node002 and be at the prompt for node002 with no password asked, and no silly kerberose failed: trying /usr/bin/rsh message given.
At this point then you can configure LAM (you may nee to download it and get it installed on your boxes)
basically it needs an arbitrary file Mynodes.txt that will contain the list of nodes you wish to launch. you type in lamboot Mynodes.txt and then it will kick back some silly error 99% of the time because something small was forgotten. you fight through those errors until it finally gives you a sucess.
Now your golden, then its just a matter of figuring out how to compile and run MPI programs with the mpiCC and mpirun. But if you got through the first gloss over then the rest is a snap.
Remember if these machines see the outside world they are naked, defenseless, and totally exploitable..
Be aware that these instructions can cause all sorts of havok and any reasonable person would just hire an expert.
Honestly I hope that this gives you a starting point. You'll still need a bunch of time with google.
GOOD LUCK!!
Storm
No, seriously, if you're setting up a cluster where your work can be batch-queued, or intend to run MPI, then Rocks http://www.rocksclusters.org/ is the way to go. It also comes with tools such as SGE (Sun Grid Engine) or OpenPBS pre-configured, Intel compilers and libraries ready for you to drop a license onto (but of course the entire GNU suite is there as well, including Ada), has more monitoring tools (plus some nice web-based interfaces) than you can shake a stick at, and runs on IA-32/AMD-64/IA-64 (Itanium). It also has a Roll to help build a tiled display wall, which would be a really cool use of a small cluster.
They're also really great guys.
On the other hand, Oscar is supposed to be good, and if you're not into the whole batch-mode thing, you can get OpenMosix up and running using http://clusterknoppix.sw.be// ClusterKnoppix, and just fire jobs off into space and let them find their own unburdened node.
But still, Rocks is really an elegant and clean way to go, plus it will scale up in case you're going to deploy a huge one of these for real after you get your feet wet.
the more accurate the calculations became, the more the concepts tended to vanish into thin air. R. S. Mulliken
This is a good reference:
http://linuxclusters.com/compute_clusters.html
Do a Google search on "Java heterogenous cluster". I have been playing around with some of them.. Nice things - you can use any machine. No lengthy build out process. Downsides - slower than dedicated, not as flexible.
Clustering means too many different things these days. I operate several clusters- but they're all so different that I can't say that any of them are the same.
I run ClamAV and Spamassassin- both very slow programs- with cexec which simply lets me farm regular unix tools across multiple (lots) of CPU servers. This lets me replace the clamscan and spamc programs with "wrappers" that use my farm. I like cexec because it doesn't make me create lists of clients and servers, but automatically load balances and fails out very nicely.
For my frontend web servers, I use fake/heartbeat and some custom proxy software for routing frontend requests to backend farms.
I haven't found a real reliable replicated directory- with one, I could use cexec as a filesystem... Maybe some day...
I worked for a linux supercomputing startup way back when. The easiest time I had was by separating the components : one big machine for storage, and lots of little diskless machines for computation.
/tmp. It's damn convenient to be able to configure all the nodes from one place whether they were online or not.
/home , /usr, etc.). You'd be amazed at the performance you can get with a well tuned NFS share... since one server can cache most of the disk access, it can even dish out files from one big high performance RAID faster than if you had bothered to give all the nodes their own drive or two.
So I'm a Debian fan, so that involves just creating one large computer (or two with redundancy using linux-ha) with a good RAID as a shared home directory. Then just install the "diskless" package. This will allow you to spawn off as many hosts that mount root off of NFS as you like. All you have to do is get the compute nodes to boot a kernel that supports nfs root (I used a single floppy, but you can do bootcds or net-boot if you're more sophisticated).
I used a Mosix kernel at the time, but I suppose OpenMosix is a better bet today. Mosix pretty much makes the entire system look like a massive SMP, so you just launch a whole lot of batch scripts on one computer, and it automatically distributes the load out to idle machines, and ships the results back to the one you started on. You just choose a node to become the master diskless-image, and then use the diskless scripts to clone it as many times as you like.
The compute nodes could have a local drive, but I just used them for swap and maybe local
The other nice thing about OpenMosix is that it's architecture agnostic, so you could conceivable join and remove nodes that were all different speeds, AMDs or Intels or maybe even 64-bit platforms in any combination. The faster processors would get the heaviest loads first, etc.
After you have this system up and running, you might start playing with more sophisticated stuff, like parallel distributed global filesystems and the like. But before that you could certainly make your NFS root server scale up by splitting it up across multiple machines (for
Anyway, it's the systems management that will get you... so I recommend using Debian, getting real cozy with aptitude, and searching the apt repository for all of the little command and monitoring thingies that will help you, like clusterssh and cfengine and nagios2 and stuff.
Burning a bunch of ClusterKnoppix CDs will pretty much get you on track with most of this, I'd imagine. Also check out the "KNOPPIX Remastering" howto so you can customize your own livecds, should you choose that path insteads of diskless nfsroot.
So that's a software approach, the hard part is really selecting, testing, and optimizing all of the hardware. The slowest component is always going to be storage (invest in lots of separate SATA cards using the Linux software RAID5 or RAID10 - reconfigure and test lots with hdparm -t and bonnie++ and format reiserfs), followed by network (gigabit NICs are cheap - you could afford separate ones for the NFS and the "real" network, though gigabit switches are still up there - Linksys and D-Link make some good 16-port ones for ~$300).
Um, if you're looking for parallel applications, povray is fun. And for a time we could sort of measure how many nodes I had up and running by monitoring my stats at distributed-net . But with OpenMosix, just about anything with lots of CPU-intensive parallel batch processing is fair game and works out of the box.
Have fun!
Last year I built a cluster using OSCAR http://oscar.openclustergroup.org/
I haven't tried any others, but OSCAR installs pretty easy. Just run the installer on the head node, and when it is done it feeds an image to each of the other computers that are a part of the cluster. It includes the ganglia monitoring tools and the apache server so you can view it.
...
Common people. It really saddens me that te only reason people can think for doing this is rendering, compiling, and coolness. Maybe, and I'm wishing more than expecting, the guy is compiling a new breed of kernel for super gaming. I think the most fun thing to do now is assume he is doing it from a gaming point of view and move into fun, spectulative hypothesising. If it doesn't help the poor guy, then at least it may give him some muc cooler ideas.
oh, wait, never mind.
I have used http://www.warewulf-cluster.org/ for my Opteron cluster. Works with new kernels and with many different distros. Seems to be under good devopelment.
This being said, for an instant trial, there are some OpenMosix LiveCDs, such as Quantian or other variants of Knoppix. Put the Quantian DVD in the 1st PC, boot, enable the remote boot option, boot the other computers over the network. Here: you have an operational cluster.
I think there may also be Rocks LiceCDs but haven't tried them. And remember your electricity bill when playing with clusters !
Non-Linux Penguins ?
What do you want to DO with it? To get one thing straight, beowulf is a distribution, and bproc/mosix/lam/mpich are ways of getting apps to communicate over a cluster. What technology you use is going to depend on the app. If the app is written for mpich, you have to use mpich. If it's written for bproc, you need to use bproc. If you're writing it, look around at the various technologies and see which one you like the most. MPI is a smallish layer above sockets that allows you to explicitly pass messages from cluster node to cluster node. Bproc allows a program to fork across the nodes of a cluster and then join back together. For just getting started (and I'm probably biased since I work for scyld), Beowulf is awesome! The latest distro is about to go beta fairly soon. It installs on top of redhat, and right after you install, you can power up the nodes and if they're set to pxe, all the nodes will come up as compute nodes. It comes with MPICH and Bproc and a few interesting demos (tachyon, a raytracer; and a fractals program) and linpack. The only bad thing about Bproc is that it has to patch the kernel. However, it works very well. I've heard bad things about OpenMosix, it does some fairly bad things like migrating file descriptors and some other silly things. The main thing is it's pretty much a hack to get threadded applications not intended to be clustered to distribute across nodes, which is just... not a good thing. Applications should be written to work in a clustered environment. Anyway, have fun!
set up an NFS share on the head node, it will simplify a whole bunch of data collection. oh, dont overwrite the home directory as your share, it will cause weird issues, make the directory something like "clustershare" it will make sure that you launch the exact same executable on all the boxes.
The LAMRSH variable needs to be set to "rsh" otherwise you get an error message.
________________
Most of this guide is if your looking to build a cluster to understand how they work, quite a bit of the software out there can make things quite a bit easier to initially set up, easier to maintain, and simpler to diagnose an issue.
I'm a fan of Warewulf, it has a reasonable memory footprint, a nice node management system. As far as diskless boxes go I've found it to be really doable. And learning how to build a bootable disk image is really gratifying (but not manditory). I havent played with "Rocks", but it seems to get rave reviews.
As far as MPI programming goes, it provides control over when and how you communicate, without totally leaving you at a bare metal interface. Systems that attempt to hide those layers may be nice, but leaves you wonderin if something wonky is making the system slowdown.
Storm
There are plenty of clustering LiveCDs out there. Hacking Knoppix* has a chapter covering the installation and use of ParallelKnoppix. PK is astonishingly easy to set up yet pleasantly powerful. I'd definitely recommend playing with a few of these CDs/DVDs. Which one is most appropriate depends on the problem you're trying to solve. There's also the slight detail of just how god-awful complicated designing and writing clustered code is, but that comes after some really fun demos. My favorite is, by far, the MPI Mandelbrot viewer covered in that chapter. It's just so damned pretty to watch.
* You probably don't need to buy the book for the content in the chapter, though.
Disclaimer: I wrote the chapter, but I don't make any money if you buy the book. So, by all means, skim it at Borders, and write a nice review of that chapter specifically on Amazon so it makes my resume look better =)
Just use Microsoft's new HPC solution, no really.
I tried it on a cluster with 22 nodes. After 2 years I threw it out. One node had one weak bit and it was impossible to find with OpenMosix, as the errors would just seemingly happen on all nodes.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
For learning purposes, you can save a lot of headaches by setting up a VWware linux setup and do snapshots so you can roll back when you do something that screws up the system (had a guy try to set up an NFS server and ended up formatting the /usr directory, how that happened, I don't know). It's also quite nice as a testing platform since the vmplayer is free. What that means, is if you have a room full of computers that never get used (as I do in my undergraduate student's computer lab), I can run the vmplayer over windows as my user and let the students login separately. This is also kinda nice so that I can do a scripted active hosts file that updates the cluster setup so that I can have 2 or 20 machines without touching a conf file myself. Surprisingly, there isn't much cpu power lost when running the VM over windows, just network lag since I do the NAT option so I don't have to register new IP's. I run MPI to drive a distributed version of MCNP (neutron transport code) on my personal cluster (Rocketcalc with extra OTS PC's as extra nodes) and do the VM for fun/testing. Other than that, look at the Gentoo HPC howto's, MPI/PVM/LAM manuals, and try to plan out your cluster before starting. It's much easier to get things to work when you know what you're trying to do beforehand than to bumble through it and say "wouldn't it be cool if..."
So far, no one has mentioned. Warewulf.
I have built three Warewulf clusters in the past year. I like how light weight and customizable WW is. It consists of a bunch of scripts that netboot/etherboot/PXE boot a custom RAM disk as your root file system from a tftp server (in my case the head node). (The smallest RAM disk we have built is around 10 MB. Everything else can be NFS mounted so each of the nodes has the capabilities of a standalong workstation.) From there you can configure it to do whatever you want. (Corollary: if you want to do it all, it may be better to start with one of the heavier cluster distributions like Rocks. I prefer my systems to be lean and mean.) By default it assumes that you are going to run "diskless" [1] but that can be overridden. I use the local disks as swap and /scratch.
[1] One of the biggest headaches in running a cluster is ensuring that each of the nodes are the same. Configuration creep is one of the banes of administering a cluster. Running diskless is a major advantage because it is easy to bring a node up to date...just reboot. It only takes 30 seconds for my nodes to reboot and be back up. (It would be about 10 seconds shorter if the BIOS would allow me to turn off the checks for Promise SATA RAID that I am not using.) Another advantage is that there is no local state so I can swap out a failed node fast with no hassle.
WW currently works best on RedHat systems, but ports for Debian and other Linux distributions are in the works. (My last two have been Debian based, but I had to do a lot of tweaking to get it right.) The community is very active and the principles are very involved in helping you be successful.
Give it a try.
For my first diskless cluster, I used the warewulf cluster solution to see one up and running. Then, I wiped the master node's disk and built one with the openmosix kernel patch etc, and used the Linux terminal services project which was really cool. the ltsp stuff made the node filesystem stuff easy to build onto. I am waiting for the openmosix team to finish up work and release the userland tools for the 2.6 kernel for my next build. here is a good how-to on LTSP+Openmosix - http://ltsp.org/contrib/ltsp-om5r3c.html
Once you get that going, you might look at PVFS2 Parallel Virtual File System. "PVFS2 stripes file data across multiple disks in different nodes in a cluster. By spreading out file data in this manner, larger files can be created, potential bandwidth is increased, and network bottlenecks are minimized."
Good Luck!!
Do yourself a favour and contact the team at Bullet Proof Linux http://www.bplinux.com/.
We needed a cluster for load balancing a typical web application, with OpenSSI http://www.openssi.org/ being the chosen SSI system. Sadly, OpenSSI is far from a "working" solution, and needs quite a bit of massaging - especially if you have newer equipment.
The guys at Bullet Proof Linux have been professional, helpful and incredibly patient in their efforts to get us going. Worth every cent.
Any they did it all remotely via ssh from the other side of the planet...
if you really want grid computing, use an out-of-the-box solution like xgrid, or a sun based solution. why linux? if not for some special OS-specific reason, sun or OSX will privide better success in the cluster realm. if you want HA, look into everything fiber with FC subsystems and brocade and heartbeat. please, this is not flame bait, just tired of the "linux everywhere" mentality. using the right tool for the job is more important...
three can keep a secret, if two are dead - benjamin franklin