Follow Slashdot blog updates by subscribing to our blog RSS feed

 


Forgot your password?
Close
typodupeerror
×
Linux Software

High Availability Clustering 45

Posted by CmdrTaco from the for-those-of-you-with-extra-hardware dept.
Christopher Cashell writes "Everyone knows about Beowulf High Performance clusters, but it's often remarked that these are impracticle for most business uses, and that High Availability Clustering is still lacking. It looks like the guys at TurboLinux are working on fixing that. First seen on freshmeat"
This discussion has been archived. No new comments can be posted.

High Availability Clustering More

High Availability Clustering

Comments Filter:

  • Re:What does all this mean? (Score:1)

    by Anonymous Coward
    High availability means two or more servers act as a single logical server, so if one node fails (i.e., root disk crash, power supply failure, etc.) the other takes over for the cluster, and services (web, nfs, dbms, etc.) continue.

    The takeover requires importing the data disks (dual initiated SCSI arrays attached to both servers), and starting the services on the surviving node.

    The heartbeat is the way each server tells the other, "I'm alive". When one server stops hearing it, it tries to take over. Sometimes, there is a serial connection that is used to force the non-surviving node to an init 0 state.

    This is the hard part, since if a NIC fails, both could assume they are in charge (called split-brain), and kill each other off, or both continue to run.

    If all works well, the failover takes from 30 seconds for a simple configuration to several minutes on a complex configuration.
  • Hang in there Ellis.

    People who set themselves up as judges tend to become arrogant SOBs with no sense of humor.

    I'm looking for a way to invert the ratings scheme that I only see post that have a 0 or less rating. I get tired of high-falutin long winded garbage and like short humorous posts better.

    Humor seems to be that last thing these self-appointed 'moderators' understand. They're too busy hiding the kids (Linux Loonies) so as not to offend the suits (don't ask me why).
  • Well, the news is that they are looking for sponsors who may also decide on the license. Would be a chance for those who want to see it GLPed, wouldn't it?
  • Excuse my gross ignornace (i know, i know.... :), but what exactly is the difference between the bewulf clusters that are all the rage at the moment and a "High Availibilty Cluster" - Jaymz, who is an ignoramus
  • Check it out, I think they are prevented from using GPL for legal reasons (government, powers that be???).

    In this case, handing MOSIX over to another team that can relicense it is the way forward.
  • Hi, I'm familar with and impressed by the eddieware project. One thing about high-availability solutions is that vendors require intensive hand-holding, enterprise 24x7 global support, and commercial technical integration.

    We're simply trying to help Linux get into the enterprise. There is no need for competition. I hope we all succeed in our own projects. If every project succeeds, Linux in general becomes stronger and this is what we really want.
  • Hi, as far as I know, the monitoring code was all developed by TurboLinux and is all original. The kernel development also diverged from Wensong's OpenSource project and future versions will continue to diverge. Since the kernel patches will be under GPL, it is possible that Wensong's project may take in the TurboLinux work. Though, that would be up to them.
  • The Beowulf clusters run in parallel and require
    special programs written for these clusters. These are primarily used in intensive computation areas such as thermal modelling.


    The clustering solution that TurboLinux has is a
    routing technology combined with monitoring tools. The routing technology allows multiple servers to act as a single server. This is the primary goal of Wensong's virtual server project. The other component is a system of monitoring utilities. These are based on heartbeat tools that send signals to the different nodes.


    The TurboLinux solution also has heartbeats for the main router node. Every node on the cluster can be both a router and a server.

  • Another thing to consider is that since all our development work on the kernel patches are and always will be under the GPL, all OpenSource projects will benefit. In version 2.x of the cluster product, we will completely rewrite the kernel modifications to drastically improve routing performance and will contribute this back to the community so that all can benefit.
  • This person's comments are not being moderated down. Check out the user's info, this person has posted 299 (!) comments in the last few weeks, and so many of them have been marked down to -1 (usually for good reasons) that their comments now default at score 0 just like an AC.
  • it looks like some other company other than Redhat is starting to give back to the linux community. Redhat has give us gnome and rpm, now TL is giving us improved clustering.
  • Apparently the only reason that they `needed' not to release it under GPL was that they wanted to hinder undesirables getting supercomputers.
    Thus they want to release it as binary only module limited to (presently) six nodes.
    I don't see the logic in that myself ... but hey - its not my code.
  • I would like to see a cluster of the new playstations, when the come out.. 128 bit RISC processor running at 300mhz. 6.4gflops!!!!
    =>
    www.playstation.com
    "Windows 98 Second Edition works and players better than ever." -Microsoft's Home page on Win98SE.
  • Why do people think Red Hat are evil? Redhat release all the things they develop under the GPL. Companies like this, who have built themselves up by selling Linux (ie little cost of R&D), then make binary shareware to put in their distros, which should be shunned.

    Almost every binary shareware distro is based on RedHat as well! Look - SuSE( Probably the worst), Caldera, TurboLinux... why do people use this stuff? So they can make more binary-only things? If you want a corporate backer to your distro, use RedHat. If you want a totally free & professional distro, use Debian. Whatever you do, *don't* support companies who are just proving that its ok to leech off the community.
  • From the page above:

    TurboLinux High Availability Cluster. Note that this is actually a repackaging of code from Wensong Zhang, who wrote it as part of the Linux Virtual Server Project which is referenced below. They do not appear to credit Wensong for his work, and resell it as "commercial" software. Are they within Wensong's license constraints (GPL?)?

    Interesting... looks like they will honor the GPL on some of it. Anyone know if the monitor and config will be based on the same GPLed material?

  • Web service is one of the easiest things to make highly available. There's very little server-side state, reads predominate over writes by a couple of orders of magnitude, etc. In fact, the TurboLinux white paper even states that they don't support any form of data consistency, so your traffic had better be _all_ reads. Yuk. They also talk about the "active router" providing "fault tolerance", indicating that they don't know the difference between fault tolerance and high availability.

    Some of the monitoring code is useful in other contexts, but they got that from somewhere else. As near as I can tell, PHT's own technical contribution is near zero.
  • There have traditionally been two streams of cluster development. One stream, represented by Beowulf, is high performance. Such clusters are actually _less_ reliable than single machines, since a failure of a single component can cause failure of an application running across the entire cluster. The other stream, represented by products such as Digital's TruClusters or Clam/IBM's HACMP/6000 (on which I used to work), is high availability. High availability in general is about reducing _total downtime_ for applications. There may be an interruption, but it should be short (long enough for takeover to occur, which is less time than a full reboot). This contrasts with actual fault tolerance, in which the occurrence of the failure is completely hidden to users or perhaps even to the application itself. Both HA and FT systems tend to become _more_ reliable as components are added, but they also tend to perform worse than the same hardware in a non-HA setup due to synchronization costs etc.
  • Not quite a fork off, more of a fork on :) We started out differently, but ended up using and modifying the kernel patch for the virtual server project. FYI we are doing most, if not all, of the development on the 2.2 patch, afaik.

    Later,

    Justin
  • Read the reply to the 'GPL versus $1,000' post for more detail, but we haven't declared closed-source yet, as the FAQ says we have not decided on Licensing for the monitoring and configuration portions yet (as it puts it, this basically covers tlclusterd which does load balancing, makes sure all the other hosts are up using more than just a ping - it actually tests the web server, and tlclusteradm, the basic admin tool.. the TurboNetCfg module that admins the CWS is open-sourced, as is the rest of the Turbo* config tools).. stay tuned, as I said, for our decision on the licensing for these tools.

    Any questions? e-mail me : justin@turbolinux.com, we're out to make people happy! :)

    Later,

    Justin
  • It's Like that, and thatz the way it is!
    F*ck commercial linux!! (well then ya dont get support than... > So?? what the X*X* should I care?? Mailing some lists would give me more support than I would expect..!)

    GreetZz

    Linux-Lover
  • Most of the customers for high-availability clusters are corporations who use them in mission-critical systems. To them $1k/node is nothing.

    As with all GPL'd software, you pay money for the intangibles-- tech support, documentation, and, in this case, value-added tools. When nodes running your stock market go down, you want to be able to pick up the phone 24x7 and get help real fast.
  • MOSIX has actually been posted to slashdot before, and was the subject of a rather intense discussion.

    They're project is a kernel module that requires kernel modifications, however, according to Linus's decision, this means they have to release the whole thing under the GPL if they decide to distribute the module. Last I heard, they hadn't decided what they were going to do about that, yet, as they wanted (needed?) to distrbitute it as binary only.
  • 1.1 What is the price of the system?

    The price is not set. However, it is anticipated to be around $1,000 dollars per node.

    1.6 What license will the software use?

    The kernel patch will be released under GPL with full source code. TurboLinux has not announced licensing for the monitoring and configuration portions of the cluster system.


    ok, GPLed!... And at $1000 a node, someone will write a GPLed monitor and config portion real quick to match it.

    Wondering why they are charging so much....

  • Yesterday btw there was an eddieware press release
    that their open source web stuff will be doing a
    real live test for the next cricket series.

    Well cool
  • There is some really good stuff on Virt Servers at http://proxy.iinchina.net/~wensong/ippfvs/ . I belive that the pacific tech distribution was a fork off of this project.
  • I'm a bit confused. OK there are many Web server very important but high avaiabilty is not only for Web.
    In the industry HA is more important for DBA, TP monitor and application server than Web? Doesn't it?

  • Re:Linux high availability (Score:3)

    by craigoda ( 7137 ) on Monday May 10, 1999 @03:47PM (#1898779) Homepage
    Hi, the monitor and config tools were developed completely seperate of any other project. Although the kernel patch by Wensong eventually came to be used in modified form in our product, it is being replaced in future versions. We tried to give Wensong appropriate credit on our main web site http://www.turbolinux.com. The credits to Wensong were there from the time we announced the product and we have a good relationship with Wensong.

    I've linked this site to our community home page.

    The statement that it is simply repackaged is incorrect. As the development continues, we'll always release the source for the kernel routing features and hopefully help the progress of the free software projects.

  • Yeah, but read the license! (Score:3)

    by Howard Roark ( 13208 ) on Monday May 10, 1999 @03:14PM (#1898780)
    Pacific HiTech is putting this out as closed source. Pretty neat. The community does all the testing and they get to keep all the source. Not very friendly.

    My computer, my way. Linux
    --
    Howard Roark, Architect

  • Linux high availability (Score:3)

    by Alan Cox ( 27532 ) on Monday May 10, 1999 @03:10PM (#1898781) Homepage
    There are a whole pile of solutions see

    http://www.henge.com/~alanr/ha/

    I guess thats a very underpublished URL 8(

  • MOSIX (Score:4)

    by michael ( 4716 ) on Monday May 10, 1999 @03:23PM (#1898782) Homepage
    Well, everybody's talking about Beowulf, about High Availability. But nobody dares to talk about MOSIX, a Linux Kernel Module developed at an Israel University. It supports things like application-transparent adaptive load balancing, memory ushering and things like that. The only problem: not free up to now. But, wait! They are currently looking for a sponsor for maintainance and further development. This sponsor(s) may also choose the license, too. Check this one: MOSIX Homepage [huji.ac.il]. Tried to post it to slashdot wo times before, but they didn't seem to like it ... :(

  • Eddieware does more.. (Score:4)

    by liquidsky ( 45979 ) on Monday May 10, 1999 @07:56PM (#1898783)
    Turbolinux only does local area clusters; and they use a very ugly, non portable, kernel patching solution.

    Eddieware does DNS load balancing (and hence isn't bottlenecked like the Linux Virtual Server Project), LAN load balancing, IP migration and admission control. In addition to linux 2.0.x and 2.2.x it works under FreeBSD and Solaris. Checkout http://www.eddieware.org [eddieware.org].

    Checkout the Eddieware press release at www.eddieware.org/txt/press990503.html [eddieware.org]. Funny how this didn't get a mention in the main slashdot articles but a closed source solution does!

  • Re:GPL versus $1,000? (Score:5)

    by TurboJustin ( 34296 ) on Monday May 10, 1999 @08:01PM (#1898784) Homepage
    As the other reply says, our main target is going to be business customers. If you can grab the kernel patch, recompile your kernel, etc.. you don't buy commercial software.. you probably also work for a company that pays you enough that maybe the time it takes you to do that costs more than the software :) The point is that our software is commercially developed and supported, we make it easier to manage and easier to implement, and take care of some of the work that would cost our customers time and money on our end..

    As far as a GPLed monitor and config, it isn't just a 'monitor and config', it is a cluster daemon that does load balancing and offers up faul-tolerance, etc.. Let's also pay attention to 'TurboLinux has not announced Licensing for the monitoring and configuration portions', this means we are looking at different licensing models, it doesn't mean we are definitely going to close the source.. stay tuned for our decision, we aren't out to chet the Linux world, we just want people to buy our stuff.

    FYI part of the config is opensourced, the module to TurboNetCfg that sets up the cluster webserver is open-sourced (the beta download is approx 120k I think, it has the kernel patch, tlcusteradm, tlcusterd, and the module for NetCfg, check it out)

    Free Software is a great thing, but so is paying the bills, it can be hard to find a balance. This is not a product that would sell in the numbers that the others will, and it requires *much* more development work than the core distribution, so it is more costly.

    If you have suggestions on any of our products, comments, etc.. let me know, my e-mail is justin@turbolinux.com, I'm a Developer Relations Associate with PHT and beleive me, we really do care :)

    Later,

    Justin

Slashdot Top Deals

Your computer account is overdrawn. Please see Big Brother.

Close