Clustering - how can I....... - VMS

This is a discussion on Clustering - how can I....... - VMS ; I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I am still old fashioned enough to describe them as Vaxen). Their purpose in life is to run WASD. I want the higher-availability of a cluster, the ease ...

+ Reply to Thread
Page 1 of 2 1 2 LastLast
Results 1 to 20 of 27

Thread: Clustering - how can I.......

  1. Clustering - how can I.......

    I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I
    am still old fashioned enough to describe them as Vaxen). Their purpose
    in life is to run WASD. I want the higher-availability of a cluster,
    the ease of management of single UAF etc.....

    So, the data source for the web server is an NFS file location stored on
    a Windows (sigh) server using RAID5. All is sweetness and light, single
    redundant data source and higher-availability web servers. Now, how can
    I get this connected to the internet and also get any inbound connection
    to hit somehting like a cluster-alias-IP address?

    All this is running behind a small router which can only direct port 80
    traffic to a single IP address. In the "good old days", I could have
    LAT running and use a cluster alias and a terminal would connect to
    whichever cluster node was available. But how can I do this with TCP/IP
    - or, can I do this with TCP/IP?

    What I want is to have all port 80 hit any cluster node - and I don't
    have any load balancing servers to fiddle with TCP/IP after it passes
    through the router.

    Any thoughts appreciated - even those that (may) say it can't be done
    without buying new and expensive comms equipment

  2. Re: Clustering - how can I.......

    On May 17, 7:23*am, Gremlin wrote:
    > Now, how can
    > I get this connected to the internet and also get any inbound connection
    > to hit somehting like a cluster-alias-IP address?


    Ummm... You would use the cluster-alias-IP address feature in the TCP/
    IP interface configuration perhaps?

    RTFM.

    $
    $ tcpip
    TCPIP> help set interface /cluster

    SET

    INTERFACE

    /CLUSTER

    /CLUSTER=host
    /NOCLUSTER

    Optional. Default: None.

    Specifies the cluster host name (alias host identifier).

    Before using this qualifier, first define the same name in the
    hosts database.

    /CLUSTER=host associates the alias host identifier with each
    interface in a cluster.

    /NOCLUSTER disables Internet cluster processing on the
    specified
    interface.

    CAUTION

    When you specify /NOCLUSTER, active communication is aborted
    for applications bound to the cluster alias name.


  3. Re: Clustering - how can I.......

    FrankS wrote:
    > Gremlin wrote:
    >> Now, how can
    >> I get this connected to the internet and also get any inbound connection
    >> to hit somehting like a cluster-alias-IP address?

    >
    > Ummm... You would use the cluster-alias-IP address feature in the TCP/
    > IP interface configuration perhaps?
    >
    > RTFM.


    And if you have a more recent VMS / TCP/IP Services version, it's called
    FailSAFE IP.

    See http://h71000.www7.hp.com/doc/83fina...#failsafe_chap

    cu,
    Martin
    --
    One OS to rule them all | Martin Vorlaender | OpenVMS rules!
    One OS to find them | work: mv@pdv-systeme.de
    One OS to bring them all | http://vms.pdv-systeme.de/users/martinv/
    And in the Darkness bind them.| home: martin.vorlaender@t-online.de

  4. Re: Clustering - how can I.......



    Gremlin wrote:
    > I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I
    > am still old fashioned enough to describe them as Vaxen). Their purpose
    > in life is to run WASD. I want the higher-availability of a cluster,
    > the ease of management of single UAF etc.....
    >
    > So, the data source for the web server is an NFS file location stored on
    > a Windows (sigh) server using RAID5. All is sweetness and light, single
    > redundant data source and higher-availability web servers. Now, how can
    > I get this connected to the internet and also get any inbound connection
    > to hit somehting like a cluster-alias-IP address?
    >
    > All this is running behind a small router which can only direct port 80
    > traffic to a single IP address. In the "good old days", I could have
    > LAT running and use a cluster alias and a terminal would connect to
    > whichever cluster node was available. But how can I do this with TCP/IP
    > - or, can I do this with TCP/IP?
    >
    > What I want is to have all port 80 hit any cluster node - and I don't
    > have any load balancing servers to fiddle with TCP/IP after it passes
    > through the router.
    >
    > Any thoughts appreciated - even those that (may) say it can't be done
    > without buying new and expensive comms equipment


    I haven't tried this yet, but it's on my TODO list:

    Using DNS to Balance Work Load
    http://h71000.www7.hp.com/doc/83fina...#toc_chapter_7

    Jeff Coffield
    www.digitalsynergyinc.com

  5. Re: Clustering - how can I.......

    On May 17, 7:23 am, Gremlin wrote:
    > I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I
    > am still old fashioned enough to describe them as Vaxen). Their purpose
    > in life is to run WASD. I want the higher-availability of a cluster,
    > the ease of management of single UAF etc.....
    >
    > So, the data source for the web server is an NFS file location stored on
    > a Windows (sigh) server using RAID5. All is sweetness and light, single
    > redundant data source and higher-availability web servers. Now, how can
    > I get this connected to the internet and also get any inbound connection
    > to hit somehting like a cluster-alias-IP address?
    >
    > All this is running behind a small router which can only direct port 80
    > traffic to a single IP address. In the "good old days", I could have
    > LAT running and use a cluster alias and a terminal would connect to
    > whichever cluster node was available. But how can I do this with TCP/IP
    > - or, can I do this with TCP/IP?
    >
    > What I want is to have all port 80 hit any cluster node - and I don't
    > have any load balancing servers to fiddle with TCP/IP after it passes
    > through the router.
    >
    > Any thoughts appreciated - even those that (may) say it can't be done
    > without buying new and expensive comms equipment


    Gremlin,

    Failsafe IP would present a single IP address for the router. That
    will provide "fail-over". It will not provide load balancing.

    To achieve load balancing on www servers, you would need to use:

    - a router that supported multiple external addresses each mapping to
    a different internal IP address
    - DNS entries that distribute the translation results over the
    different external addresses.

    - Bob Gezelter, http://www.rlgsc.com

  6. Re: Clustering - how can I.......

    On May 17, 8:21*am, "Martin Vorlaender" wrote:
    > And if you have a more recent VMS / TCP/IP Services version, it's called
    > FailSAFE IP.


    Not really.

    Failsafe IP allows a system with multiple interfaces (that is, network
    adapters) to recover from the failure of any single adapter.

    The original poster was very specific in asking how to get clustered
    systems to share an IP address. That's not the same as multiple
    interfaces on a single system.

  7. Re: Clustering - how can I.......

    On May 17, 8:58*am, Bob Gezelter wrote:
    > To achieve load balancing on www servers, you would need to use:
    > - a router that supported multiple external addresses each mapping to
    > a different internal IP address
    > - DNS entries that distribute the translation results over the
    > different external addresses.


    That's not really load balancing.

    Round-robin DNS just gives out a different address for each
    translation request. It doesn't take into account the actual load on
    the system(s), nor does it account for DNS caching on the client.
    Caching would prevent repetitive requests for the same translation,
    thereby putting an unbalanced load on the target hosts because the
    same IP address would be used until the TTL expired.

  8. Re: Clustering - how can I.......

    On May 17, 3:29 pm, FrankS wrote:
    > On May 17, 8:58 am, Bob Gezelter wrote:
    >
    > > To achieve load balancing on www servers, you would need to use:
    > > - a router that supported multiple external addresses each mapping to
    > > a different internal IP address
    > > - DNS entries that distribute the translation results over the
    > > different external addresses.

    >
    > That's not really load balancing.
    >
    > Round-robin DNS just gives out a different address for each
    > translation request. It doesn't take into account the actual load on
    > the system(s), nor does it account for DNS caching on the client.
    > Caching would prevent repetitive requests for the same translation,
    > thereby putting an unbalanced load on the target hosts because the
    > same IP address would be used until the TTL expired.


    DNS load balancing is about as simple as you can get, but is often too
    simple to be much use, as you point out.

    I can't comment on this subject from recent personal experience, but
    The Fine Manual referred to earlier in the thread talks about a "load
    broker", which if it works as usually intended, would seem to be at
    least as adequate as LAT-style load balancing ever was. Active load
    balancing at connect time rather than on a per-request basis can have
    a few snags, which can make it not worth the effort, but...

    Anybody out there got real-life experience of the load broker and care
    to share it?

    regards
    John

  9. Re: Clustering - how can I.......

    In article , Gremlin
    writes:

    > I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I
    > am still old fashioned enough to describe them as Vaxen). Their purpose
    > in life is to run WASD. I want the higher-availability of a cluster,
    > the ease of management of single UAF etc.....
    >
    > So, the data source for the web server is an NFS file location stored on
    > a Windows (sigh) server using RAID5.


    WHY?!?!?!?!?!

    > All is sweetness and light, single
    > redundant data source and higher-availability web servers. Now, how can
    > I get this connected to the internet and also get any inbound connection
    > to hit somehting like a cluster-alias-IP address?


    Just set up the cluster-alias IP address and have the outside world use
    that. If you are using some sort of NAT/PAT router, have it direct
    incoming stuff (at least for your WWW port) to the cluster-alias
    address.

    > All this is running behind a small router which can only direct port 80
    > traffic to a single IP address. In the "good old days", I could have
    > LAT running and use a cluster alias and a terminal would connect to
    > whichever cluster node was available. But how can I do this with TCP/IP
    > - or, can I do this with TCP/IP?


    One and only one machine has the cluster-alias address at any given
    time. It provides failover, not load-balancing.

    I don't know much about WASD, but it might be possible to have scripts
    etc be executed on other nodes (or on some load-balanced DECnet-alias
    node).


  10. Re: Clustering - how can I.......

    In article
    ,
    FrankS writes:

    > On May 17, 8:21=A0am, "Martin Vorlaender" wrote:
    > > And if you have a more recent VMS / TCP/IP Services version, it's called
    > > FailSAFE IP.

    >
    > Not really.
    >
    > Failsafe IP allows a system with multiple interfaces (that is, network
    > adapters) to recover from the failure of any single adapter.


    Actually, the old cluster-alias IP stuff is now subsumed into the
    failsafe-IP stuff. Confusing, yes, but that's the way it is.

    Also, note that, as far as I know, there was no time when one had both,
    to make for an easy transition. There weren't even obsolete commands
    for backwards compatibility.


  11. Re: Clustering - how can I.......

    What could be done is an RTR like system.

    system 1 has THE IP that the router sends all HTTP transactions to.
    system 1 then forwards the transaction to the least busy of the other 2
    machines.

    But you end up with a lot of overhead.

    In the end, if you have a simple router, it is likely you'll have a
    simple DSL line to feed your cluster, and chances are that even an all
    mightly microvax II would have no problem service HTTP requests on that
    line.


    The way Failsafe IP works: when node X wants to take over the failsafe
    ip 10.0.0.50, it sends a gratuitous ARP broadcast, hoping all machines
    on the LAN will update their ARP table so that the next packet they need
    to send to 10.0.0.50 gets sent to the new ethernet address.

    (This happens to not work on my router, but it is just a netgear)

    You can't really dynamically switch this easily. Consider some remote
    user sending you a large picture as part of an HTTP request. This will
    consist of a large number of TCPIP packets. If you change the ARP entry
    midway, it means that the new host will start receiving the second half
    of the transaction without any idea of the first half, and of course,
    with no context on that TCPIP link.


  12. Re: Clustering - how can I.......

    Phillip Helbig---remove CLOTHES to reply wrote:

    > Actually, the old cluster-alias IP stuff is now subsumed into the
    > failsafe-IP stuff. Confusing, yes, but that's the way it is.
    >
    > Also, note that, as far as I know, there was no time when one had both,
    > to make for an easy transition. There weren't even obsolete commands
    > for backwards compatibility.


    The functionality was preserved.

    On Alpha:
    SYS$COMMON:[SYS$STARTUP]TCPIP$SYSTARTUP.COM

    $IF thisnode .eqs. "CHAIN"
    $then
    $! create the cluster alias
    $ ifconfig we0 alias 10.0.0.50/16 ipmtu 1500
    $endif
    $!
    $!

    WE0: flags=c43
    Standby addresses:
    inet 10.0.0.50 netmask ffff0000 broadcast 10.0.255.255
    (Active on VELO WE0)

    *inet 10.0.0.11 netmask ffff0000 broadcast 10.0.255.255 ipmtu 1500

    -----------------------


    This coexists with node VELO, a VAX with TCPIP 5.3:

    CPIP> show int ze0/full
    Interface: ZE0
    IP_Addr: 10.0.0.7 NETWRK: 255.255.0.0 BRDCST: 10.0.255.255
    Cluster Impersonator
    C_Addr: 10.0.0.50 C_NETWRK: 255.255.0.0 C_BRDCST: 10.0.255.255
    Ethernet_Addr: AA-00-04-00-07-04 MTU: 1500
    Flags: UP BRDCST RUN MCAST SMPX
    RECEIVE SEND
    Packets 40782182 52651870
    Errors 0 0
    Collisions: 0


    The "new" interface also makes it more obvious on how to switch the
    alias around. If CHAIN had the alias and I wanted it to move to VELO, on
    chain, I would:

    ifconfig WE0 -alias 10.0.0.50
    and then:
    ifconfig we0 alias 10.0.0.50/16 ipmtu 1500
    to put the al;ias back, this time, in standby mode because meanwhile
    VELO will have taken over.

    On the TCPIP interface, forcing the alias to swicth nodes is not so obvious.

    Yeah, the change in commands is not "upward" compatible, but the
    functionality has been preserved.

  13. Re: Clustering - how can I.......

    FrankS wrote:
    > "Martin Vorlaender" wrote:
    >> And if you have a more recent VMS / TCP/IP Services version, it's called
    >> FailSAFE IP.

    >
    > Not really.
    >
    > Failsafe IP allows a system with multiple interfaces (that is, network
    > adapters) to recover from the failure of any single adapter.
    >
    > The original poster was very specific in asking how to get clustered
    > systems to share an IP address. That's not the same as multiple
    > interfaces on a single system.


    I suggest you read the excellent article by Matt Muggeridge in the
    VMS Tech Journal #2,
    http://h71000.www7.hp.com/openvms/jo...les/tcpip.html

    The interfaces failing over don't have to be on one machine. What
    you're referring to is called LAN Failover.

    cu,
    Martin
    --
    One OS to rule them all | Martin Vorlaender | OpenVMS rules!
    One OS to find them | work: mv@pdv-systeme.de
    One OS to bring them all | http://vms.pdv-systeme.de/users/martinv/
    And in the Darkness bind them.| home: martin.vorlaender@t-online.de

  14. Re: Clustering - how can I.......

    Phillip Helbig---remove CLOTHES to reply wrote:
    > In article , Gremlin
    > writes:
    >
    >> I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I
    >> am still old fashioned enough to describe them as Vaxen). Their purpose
    >> in life is to run WASD. I want the higher-availability of a cluster,
    >> the ease of management of single UAF etc.....
    >>
    >> So, the data source for the web server is an NFS file location stored on
    >> a Windows (sigh) server using RAID5.

    >
    > WHY?!?!?!?!?!




    Because the cluster is of DS10Ls, using local IDE storage and the only
    simple, cheap way I can have a single source for the web server (I
    think) is to have it come from an NFS source. Why Windows? I have it
    running already and I am happy to not add another OS to the mix....

  15. Re: Clustering - how can I.......

    In article , Gremlin writes:
    >Phillip Helbig---remove CLOTHES to reply wrote:
    >> In article , Gremlin
    >> writes:
    >>
    >>> I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I
    >>> am still old fashioned enough to describe them as Vaxen). Their purpose
    >>> in life is to run WASD. I want the higher-availability of a cluster,
    >>> the ease of management of single UAF etc.....
    >>>
    >>> So, the data source for the web server is an NFS file location stored on
    >>> a Windows (sigh) server using RAID5.

    >>
    >> WHY?!?!?!?!?!

    >
    >
    >
    >Because the cluster is of DS10Ls, using local IDE storage and the only
    >simple, cheap way I can have a single source for the web server (I
    >think) is to have it come from an NFS source. Why Windows? I have it
    >running already and I am happy to not add another OS to the mix....


    I don't know whether you're in hobbyist land or not, but if you are (so
    license costs aren't an issue) you could configure your DS10s to MSCP serve
    their IDE disks (which is, like, one command), then use host-based volume
    shadowing across those disks. (You can have three members in a volume set,
    so each of your servers could actually be running from local disk.)

    Is your PesonalAlpha an emulator? I think it could just NFS serve whatever
    disk image you give it, and the underlying disk can be whatever.

    -- Alan



  16. Re: Clustering - how can I.......

    In article <00A79BBB.463223B7@SSRL.SLAC.STANFORD.EDU>, winston@SSRL.SLAC.STANFORD.EDU (Alan Winston - SSRL Central Computing) writes:
    >In article , Gremlin writes:
    >>Phillip Helbig---remove CLOTHES to reply wrote:
    >>> In article , Gremlin
    >>> writes:
    >>>
    >>>> I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I
    >>>> am still old fashioned enough to describe them as Vaxen). Their purpose
    >>>> in life is to run WASD. I want the higher-availability of a cluster,
    >>>> the ease of management of single UAF etc.....
    >>>>
    >>>> So, the data source for the web server is an NFS file location stored on
    >>>> a Windows (sigh) server using RAID5.
    >>>
    >>> WHY?!?!?!?!?!

    >>
    >>
    >>
    >>Because the cluster is of DS10Ls, using local IDE storage and the only
    >>simple, cheap way I can have a single source for the web server (I
    >>think) is to have it come from an NFS source. Why Windows? I have it
    >>running already and I am happy to not add another OS to the mix....

    >
    >I don't know whether you're in hobbyist land or not, but if you are (so
    >license costs aren't an issue) you could configure your DS10s to MSCP serve
    >their IDE disks (which is, like, one command), then use host-based volume
    >shadowing across those disks. (You can have three members in a volume set,
    >so each of your servers could actually be running from local disk.)
    >
    >Is your PesonalAlpha an emulator? I think it could just NFS serve whatever
    >disk image you give it, and the underlying disk can be whatever.



    Dang, I meant "just MSCP serve", not "just NFS serve". (Got distracted in the
    middle of the post and lost my train of thought.)

    -- Alan
    >


  17. Re: Clustering - how can I.......

    In article <482f4d24$0$7236$c3e8da3@news.astraweb.com>, JF Mezei
    writes:

    > Phillip Helbig---remove CLOTHES to reply wrote:
    >
    > > Actually, the old cluster-alias IP stuff is now subsumed into the
    > > failsafe-IP stuff. Confusing, yes, but that's the way it is.
    > >
    > > Also, note that, as far as I know, there was no time when one had both,
    > > to make for an easy transition. There weren't even obsolete commands
    > > for backwards compatibility.

    >
    > The functionality was preserved.


    Yes. However, procedures would break after an upgrade if they were not
    correspondingly changed. And one couldn't change them beforehand,
    because as I said there was no time when both sets of commands worked.


  18. Re: Clustering - how can I.......

    In article , Gremlin
    writes:

    > Phillip Helbig---remove CLOTHES to reply wrote:
    > > In article , Gremlin
    > > writes:
    > >
    > >> I have 3 Vaxen clustered (alright, 2 alphas and a PersonalAlpha, but I
    > >> am still old fashioned enough to describe them as Vaxen). Their purpose
    > >> in life is to run WASD. I want the higher-availability of a cluster,
    > >> the ease of management of single UAF etc.....
    > >>
    > >> So, the data source for the web server is an NFS file location stored on
    > >> a Windows (sigh) server using RAID5.

    > >
    > > WHY?!?!?!?!?!

    >
    >
    >
    > Because the cluster is of DS10Ls, using local IDE storage and the only
    > simple, cheap way I can have a single source for the web server (I
    > think) is to have it come from an NFS source. Why Windows? I have it
    > running already and I am happy to not add another OS to the mix....


    You're talking about a cluster. Have the "single source" be a shadow
    set consisting of disks on 2 or 3 cluster nodes. I admit I have no
    experience with IDE storage---will HBVS work there?


  19. Re: Clustering - how can I.......

    In article <00A79BBB.463223B7@SSRL.SLAC.STANFORD.EDU>,
    winston@SSRL.SLAC.STANFORD.EDU (Alan Winston - SSRL Central Computing)
    writes:

    > I don't know whether you're in hobbyist land or not, but if you are (so
    > license costs aren't an issue) you could configure your DS10s to MSCP serve
    > their IDE disks (which is, like, one command), then use host-based volume
    > shadowing across those disks. (You can have three members in a volume set,
    > so each of your servers could actually be running from local disk.)


    That should be "shadow set", not "volume set". A volume set is a
    different beast. (Yes, volume sets can be combined with shadow sets.
    In general, in these days of big disks, no-one needs volume sets
    anymore.)


  20. Re: Clustering - how can I.......

    In article , helbig@astro.multiCLOTHESvax.de (Phillip Helbig---remove CLOTHES to reply) writes:
    >In article <00A79BBB.463223B7@SSRL.SLAC.STANFORD.EDU>,
    >winston@SSRL.SLAC.STANFORD.EDU (Alan Winston - SSRL Central Computing)
    >writes:
    >
    >> I don't know whether you're in hobbyist land or not, but if you are (so
    >> license costs aren't an issue) you could configure your DS10s to MSCP serve
    >> their IDE disks (which is, like, one command), then use host-based volume
    >> shadowing across those disks. (You can have three members in a volume set,
    >> so each of your servers could actually be running from local disk.)

    >
    >That should be "shadow set", not "volume set". A volume set is a
    >different beast. (Yes, volume sets can be combined with shadow sets.
    >In general, in these days of big disks, no-one needs volume sets
    >anymore.)
    >


    You're right, of course. I was sitting at home composing this post and got
    distracted by a conversation with my housemate, and more than one error
    crept in. Should have been more careful.

    -- Alan (who happily uses HBVS and now with EVA never need to muck with a
    volume set ever again).

+ Reply to Thread
Page 1 of 2 1 2 LastLast