[Linux-cluster] GFS-Cluster and Virtualization

Mike Fedyk mfedyk at matchmail.com
Thu Mar 17 19:38:03 UTC 2005


Theoretically, you could use GFS to let your VMs running Linux access 
the same storage on one system also.  It depends on whether VMWare will 
allow the separate VMs use the same storage device for GFS.

Also remember that with clustering like GFS, your weakest link is your 
downfall.  Most likely you are implementing virtualization to mitigate 
the problems security compromises can cause.  Also realize that if the 
rootkits start attacking GFS, or cause locking problems with the 
module(s) they insert, there could be corruption in that instance.  
Someone please correct me if I am wrong.

If you are looking for load leveling, you might want to take a look at 
OpenSSI (which uses GFS and/or CFS) and (in the future) XEN.  From what 
I see, OpenSSI already can move processes from one node to another for 
load leveling, and makes allowances for the "home node" of a process to 
fail without losing service from that process.  XEN has future plans to 
move VMs between nodes, but does not do it yet AFAIK.

Mike

Thomas Liesner wrote:

>Hi all,
>
>i'm currently doing some kind of research about cluster-technology for
>io-intensive applications like file- and printserver in prepress-
>eviroments. As a total newby to cluster-concepts i was wondering if i
>could do something like the following with gfs (or other cluster
>technologies):
>
>Several servers acting as one "superserver" hosting some virtualization
>technology like VMWare. All cluster nodes would use the same shared
>storage connected via fibrechannel to a fc-switch (This is exactly what
>gfs is doing, right?) The virtual machines on this host would run
>standard linux distributions like RedHat ES 4. Especially charming would
>be the ability to grant storage and cpu needs between the  virual
>machines "on demand".
>
>1. Is it possible?
>2. Would i run into io-problems (locking etc)?
>3. Where could i find concepts of such installations with benchmarks and
>reliability tests?
>
>Thanks for any help,
>./Tom
>  
>




More information about the Linux-cluster mailing list