• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Two ESXi Servers

W

WizBomb

Limp Gawd
Joined
Dec 5, 2001
Messages
168
Is it possible to cluster two ESXi servers (Two Dell PowerEdge 2650 Servers) together to share CPU/memory?
 
If you're thinking about having one VM split across on two servers, then no.
 
Basically the solution you're looking for isn't virtualization. You'll want to look into clustering for high performance computing or a server farm of some sort. Probably want to starting looking at Beowul clustering or Microsoft HPC solutions.
 
I was just about to start a thread about this.

What about creating like a Windws Cluster, or Linux Beowulf Cluster and install VMware workstation on it?
 
What application workload are you looking at this for? This can be accomplished through active/active clusters, it just depends on if the application is designed to handle the workloads this way.
 
Virtualization is for using unused resources, not combining multiple servers. Workstation isn't built to be clustered - an application has to be designed for active/active clustering.
 
Im running Folding@home

I was just thinking of doing it....just to do it.
The application itself is multithreaded so theres no issues there.
I have server2008 on an i7 and it works across all 8 cores very well.

I am thinking I can cluster all my boxen and run windows HPC server and just run the applicaiton on it across all my boxes
 
WizBomb said:
Guess my final question is, any software soultions support this from Vmware or other vendors?
Click to expand...

Sure - but not a virtualized system like you're thinking. IBM makes some supercomputing clusters that run AIX, and you get true hardware virt that way... Beowulf clusters, etc are also out there... But it's not gonna work quite like you expect.
 
IBM x3850 M2 = 4x Socket 771, 32RAM Slots (up to 256GB of RAM), Up to 4x 2.5" SAS 15k, 6x PCI-E x8, 2x 1400w PSUs.

Add the backend expansion option, and you can scale this box up to 4x chassis. All CPU, I/O expansions, disk, and CPU are seen by the IBM x4 (Hurricane) chipset as a single entity.

I'm sure that there are other machines from other vendors (on the open systems side...read x86) that can do the same.

As for the AIX stuff, all of that starts with System P boxes. They can (in the very large systems) go all the way up to 448 CPUs (sockets) in a single rack, and each CPU can be up to a Quad core, currently with Power 6/6+ CPUs. Power 7 will double that density. IBMs implementation on system P is as such:

HMC or IVM = Hardware Management Console (this is actually a System X Server with some specific software on it to manage the System P boxes. HMC can manage multiple System P boxes. The IVM is the Integrated Virtualization Manager, and is only capable of managing the machine that it is running on. The difference between an HMC and an IVM is that an IVM is only available on the entry level boxes, and presents an interface that's similar to the HMC, but is only available through a web page interface.
>
Hardware boxes (possibly multiple with backend stacking connections)
>
Licensed VIO Server (aggregates all CPU, RAM, I/O adapters, disk, networking in the stack of boxes +
PowerVM (this is the Type 1 Hypervisor for System P, and it leverages the CPUs capabilities to partition the system all the way down to 1/100th of a CPU)
>
DLPARs of various OSs (DLPAR = Dynamic Logical Partiion, a partition that can have memory, CPU, I/O and disk added to it, while the partition is running, with no downtime. This also supports processor deallocation in the event of a pre-emptive processor failure detection algorithm). The DLPARs in this case would be equivalent to a VM inside of ESX or ESXi. A DLPAR is a self-contained OS instance with a workload (that you create, just like a VMWare VM). Power Systems can run: iOS, AIX, and Linux x86. All AIX versions maintain binary compatibility with each successive release, so hardware platform migration, or virtualization of a workload is not an issue as a customer moves to a new hardware solution. Linux x86 applications can run without code or kernel mods on PowerVM DLPARs, or you can get Linux on Power distributions from your preferred software vendor.
>
WPARS = Workload Partitions. WPARs would be similar to haveing a VMWare ESX based Server VM, which ran SQL + IIS + Sharepoint. A WPAR (in this example) would be either SQL, IIS, or Sharepoint. It's an encapsulation at the application level. If you have PowerVM Enterprise, you can move this WPAR between DLPARs, and in between physical systems (either inside of the stack, or to another hardware compatible machine being managed by the HMC).
 
and you can combine racks of the system P's - we did it at NCAR :) IIRC, the Blue Sky cluster had 2000 processors total, or so?
 
Yea i was thinking of trying to aggregae resources using consumer products.
 
lopoetve said:
and you can combine racks of the system P's - we did it at NCAR :) IIRC, the Blue Sky cluster had 2000 processors total, or so?
Click to expand...

Indeed. The fastest supercomputer right now (which our company has permanent contracted positions to operate and maintain) is Roadrunner, which uses a combination of IBM x3755 (8 Core Opteron 8xxx series nodes, used for doing the heavy lifting) and IBM BladeCenter H chassis with Cell processor Blades (used for farming out the jobs to the Opteron nodes).

The next fasted Supercomputer, which is to be called Zia will be based on Power7 architecture, and will have 8 cores per socket, operating at 5.0GHz/core or higher. It is expected to hit at or close to 10 Petaflops. Roadrunner has reached 1.3 Petaflops (officially), so far.
 
nitrobass24 said:
Yea i was thinking of trying to aggregae resources using consumer products.
Click to expand...

all platforms that support this currently are on the high end of the open systems portfolios for many vendors. most have boxes that do it, but it has to be done at the chipset level, which means that only the vendors capable of developing their own chipsets can afford the R&D costs associated with an endeavor such as this.

to be honest, I'm sure most vendors just don't see the point of trying to make an x86 system hardware clustered, as clustering through at the OS (at least, clustering for hardware resource aggregation) level has a lot of overhead associated with it. and, honestly, most x86 systems were never designed to do this, and don't have a lot of horsepower to accommodate it. I know they're powerful systems to people that think quad core is awesome, and in the traditional thinking of computers as only desktop systems, but int he grand scheme of powerful systems, open systems are at the bottom of the performance chart.

As a point of reference, a decently configured IBM x3950M2 setup (which means at least 2x x3850 M2s in a stack), say with:

3years 24x7 hardware support
4x Intel quad core xeons
128GB of RAM
4x2.5" 15k RPM 146GB SAS in RAID-10
2x QLogic PCI-E 2port FC HBAs
2x Intel Server Pro 1000 GbE PCI-E adapters
system interconnect cable

going to run around ~$55k/ea (give or take $5k), and that's with no software licenses, at all. Nevermind that you'll need FC switches, a FC RAID Controller, disk shelves, disks, support on those, etc.
 
You must log in or register to reply here.
Back
Top