Max number of partitions on IBM Power Systems
PowerVM was enhanced in April 2011 to support up to 320 partitions on the Power 750, 640 partitions on the Power 770 and 780 and up to 1000 partitions on the Power 795.
Oracle has brought true live migration of workloads to platforms using the SPARC T series of processors, by tweaking the VM Server for SPARC server hypervisor formerly known as logical domains – LDoms, for short. With Oracle VM Server for SPARC 2.1, a system can be carved up into as many as 128 LDoms. These LDoms can span many …
Sun/Oracle have been playing catchup on the LPAR capability from IBM for ages; this almost brings them to parity on features and certainly up to the stage where the feature set covers most of what you need. Live migration is one of those things people expect their virtualisation layer to do; Containers can't do it yet and I've not heard of it even being on the roadmap; the technical difficulties of such a migration are probably sufficiently complex as to be impossible...
Now, if they can provide LDOMs on something with decent single thread performance, it'll really give them a boost. If T4 can deliver that, it'll give Oracle a boost in the hardware arena.
IBM has had this function since Power6 in 2007...it takes customers about two years to test and trust this kind of technology. So while IBM's mobility is pervasive as it has been in the market for 4+ years, Oracle is just now in alpha/beta/prototype stage.
Per the comment about max partitions supported....read the Sun recommendations. Sun does not recommend putting more than one partition on a chip. The T-chip has such a small cache that the trashing is intolerable. Yes they dont timeslice, but wow that is a small cache to try to put more than one virtual machine per chip let alone per core.
I agree that a feature like this takes time to test, but from my understanding no one really uses this feature on IBM HW, as all of requirements to make it work properly are next to impossible to meet. Correct me if I am wrong... this is what I hear from friends that use Power systems (I do not).
Regarding the cache comment on the number of partitions... bull. Sun nor Oracle make any such recommendation based on cache size. SPARC does not have to have as much cache as Power since it is CMT and can get work done even while a thread is waiting on a memory load. Power will stall all threads on a cache miss, which will happen regardless of how much cache you have. When you slice up a CPU, like IBM does, you WILL have a lot of cache misses as each partition will be doing different things.
Well, I use the IBM Hardware and Software Sales Manual to figure out what is there and what is not there yet.
Check it out:
IBM PowerVM V2.2 contains the following enhancements:
Role Based Access Control (RBAC)
Support for up to 80 virtual processors on Power 710 and 720
Support for up to 160 virtual processors on Power 730, 740, 750, 770, and 780
Support for up to 254 virtual partitions on Power 795
Support for Concurrent Add of VLANs
PowerVM support for sub-chip per-core licensing on Power 710, 720, 730, and 740
I know IBM was working on extending PowerVM, but I have yet to see these capabilities in an announcement letter. Maybe they were rolled up in a Technology Release?
TPM, you can run up to 1000 LPARs on Power795 box. The limit of 254 applies to POWER6.
Besides, LDOMs are still far behind LPARs. Look at the limitation of LDOM migration: the same cpu frequency and type on both machines, it requires lots of manual pre-configuration before migration.
I'm not surprised TPM compares only to AIX, but this capability is all already available with Integrity Virtual Machines across their range of Integrity servers (last and current generation), not just the entry-level webservers the T-series are, and can do not just hp-ux but also OpenVMS virtualised servers (and Windows and Linux if you really have to). Snoreacle is still lagging waaaaaay behind.
As to the differing maximum number of Pee-series LPARs quoted, I think the difference is between what IBM tell their customers in the sales presentations (the higher numbers) versus the reality of what Global Services may actually get working (the IBM internal figures TPM quotes).
/Hi-ho, hi-ho, it's off to troll we go!
Interesting as all the numbers being thrown around about maximum number of virtualised servers, etc, the reality is no-one is going to use them because the resulting instances are too puny for a task you'd throw at UNIX rather than put on a cheaper VMware/Xen/HyperV instance. As an example, I can remember playing with the first gen Integrity Virtual Machines and creating hp-ux VMs with as little as 128MB of memory, which was pretty pointless. With the latest release, hp recommends at least 512MB of memory per hp-ux VM, but I'd laugh at that, we like to give them at least 2GB and pref 4-8GB for real oomph. If they're going to run a proper database instance then 16GB and up. Of all the Pee-series servers we have, none are running more than a dozen LPARs for exactly the same reasons. Sorry, no Slowaris servers left to compare with!
Even funnier, why is it a thread on a Slowaris article dominated by IBM trolls all talking up LPARs? Have the Sunshiners all rasied the white flag?
But I will acknowledge that LDOMS's are largely useless for us as far as virtualization goes, we have a few messaging app's and web servers which could use it, but it's hardly worth the effort when the numbers we run are so low.
We use zones extensively to virtualize our Oracle DB's on the M-Series servers, and it works extremely well, as 90% of our Sun servers run Oracle, we don't really need anything else..
I will take another look at LDOM's once the T4 come's out.
Biting the hand that feeds IT © 1998–2021