Sep 16, 2014
Hitachi’s Intel Xeon based Compute Blade servers (CB 2000 and CB 500) feature an embedded logical partitioning (LPAR) capability implemented in the blade server firmware, based on technology originally developed for Hitachi mainframe and Unix systems to provide platform partitioning for reliability and quality of service for mission critical workloads.
When used in a converged solution like the Hitachi Data Systems Unified Compute Platform (UCP), LPARs can be used to consolidate workloads such as scale out SAP or multiple Oracle RAC and non-RAC applications, reducing hardware and software licensing costs and complexity. Now with the new Intel E5v3 processor (aka Haswell), Hitachi x86 Compute Blades can also support the consolidation or “nesting” of Virtual Machine Managers (VMM) like VMware, Hyper-V, and KVM, on LPARs in a single or SMP blade system.
Intel’s new Xeon E5v3 processor has a new feature aimed at improving performance of nested Virtual Machine Managers – the ability for a root VMM to support guest VMMs. Although this has been technically possible in previous Xeon generations, it was limited in practicality because of the high performance overhead involved when the host VMM and the guest VMM contended for VM Control Structure (VMCS) for reads and writes. Running a VMM on an LPAR had the same effect on performance as nested VMMs due to contention for VMCS access.
The new Xeon E5v3 Intel has removed this performance barrier by virtualizing the VM Control Structure (VMCS), creating a VMCS Shadow, which reduces the virtualization latencies – a feature which was originally proposed by, and then developed in cooperation with, Hitachi engineers.
With the new Xeon E5v3, we can now run VMMs, like VMware, Microsoft Hyper-V, and KVM, on separate LPARs in a Hitachi Unified Converged Platform (UCP) with minimal performance impact. The synergy of VMM and LPARs enables the VMM to hide the physical hardware and emulate a virtual machine OS on top of it to provide superior flexibility and management, while the LPAR provides partitioning of hardware resources and exposes the subset of hardware components to each OS in the system. The LPAR provides safe multi-tenancy for VMMs that share the same physical server. This is similar to the partitioning that Hitachi provides in their storage virtualization systems, the HUS VM, VSP, and VSP G1000.
The hardware partitioning of LPARs provides these benefits in shared VMM environments and cloud services.
1. In multi-tenancy environments, LPARs allow application isolation and high security by running each client’s VMM in its own hardware LPAR.
2. LPARs also make it possible to run different versions of VMM – or even different vendors’ VMMs – on the same hardware for migration, development or production.
3. Customers can mix VMMs with bare metal OSs such as Windows Server OS or Linux on the same hardware for an optimum combination of performance and flexibility.
While virtualization enables the pooling of resources for greater utilization and efficiency, virtualization must be accompanied by dynamic partitioning to isolate the failure in one partition from affecting users in other partitions. Partitioning also prevents the user of one partition from stealing resources from another partition, and prevents the leakage of data and escalation of management privileges. LPARs should be a key requirement for server virtualization with the new Intel E5v3 processors.
General availability (shipping) of the new E5v3 based version of our CB520H dual-socket blade for the CB500 blade server will be in October, but orders may be placed now. It’s known as the CB520Hv3 blade.
Hear our VP of Storage Product Management Roberto Basilio’s interview on Intel’s ChipChat with Allyson Klein. Roberto explains what Intel’s announcement of the E5v3 means for our customers. You can also see my blog on Intel’s Community site.