Question about IBM's Virtualizing history

Joined
Dec 10, 2006
Messages
540
I just was surfing around and I found stuff saying that IBM has had a history of virtualizing for the past 30-40 years. Can someone tell me what they were doing in all that time in terms of virtualization before all the current day virtual stuff?
 
Mainframes. Multiple instances on mainframes. People still do it..look up zLinux.
 
NetJunkie got it. Primarily it's a Mainframe and i Series thing.

AIX only recently got it. DLPAR, LPAR, microLPAR? Those are NOT virtualization. They're partitioning. WPARs are virtualization.
 
AreEss - I understand the process in basic, but can you elaborate on how those do partitioning? I've always been curious about the process behind the curtain :)
 
AreEss - I understand the process in basic, but can you elaborate on how those do partitioning? I've always been curious about the process behind the curtain :)

See, now we're talking my area of expertise.

The way partitioning works is in conjunction with something called a Service Processor or Flexible Service Processor. (SP or FSP.) The SP dates back to the 603e/604e RS/6000's, but the modern implementation only dates back to about POWER3/POWER4.

The Service Processor works in conjunction with the HMC (Hardware Management Console) to electrically and/or logically divide system resources including but not limited to processors or cores, DIMM sockets or memory groups, expansion chassis, expansion slots, RIO2 or GX2 ports, and/or individual card ports including but not limited to SCSI, Ethernet, etcetera.

It's possible to take a 32 core P570 and divide it into 32 single-core LPARs. It's also possible using AIX 6.1 to create a shared processor pool and create 64 32-core LPARs. Performance will SUCK, but it's possible.

Now, in conjunction with LPARs you can use Virtual I/O Servers. These are special LPARs running a special version of AIX. You assign them resources like any standard LPAR, except that they require far less, and pass virtualized storage and networking up into standard LPARs. These also can be used to provide fault tolerance and HA by allowing you to fail resources across multiple VIO Servers - upgraded to multiple VIO Servers on multiple independent chassis when adding PowerVM Enterprise to the mix. They also allow you to increase utilization and decrease port count by combining multiple low-traffic applications into a single VIO Server.

Obviously, I'm oversimplifying here and there, but that's pretty much the super-short crash course. :)
 
to be fair, WPARs are more akin to vApps than anything else. WPAR is a containered application load, which still requires an underlying OS (LPAR or DLPAR) to function as a host OS. However, the WPAR can be moved in between hosts (provided that you have PowerVM Enterprise, multiple hosts, and supporting versions of AIX in LPARs or DLARs on each host), kind of like VMotion, except that you're just moving the application.

AreEss got everything else right about the electrical partitioning of the system. It is worth mentioning, and that with PowerVM, Aix 5.3 or newer, and a Power 5 CPU, or newer, you can go down to 1/10th of a physical CPU being assigned to an LPAR. Hence, on a 32core P5-570 (note that Power 6 based machines lack a P6 designation, and are just listed as P-###), you could do the following (with proper licensing):

32 cores x 10 (micro partitions /core) = 320 LPARs
 
Back
Top