Really cool.
In collaborative tests conducted on the White Rose Grid cluster at York, RepliCator’s prototype Virtual Provisioning Optimizer capability and VMware software were employed to create a virtual container within the cluster's native 64-bit Linux environment. Within this virtual container, RepliCator then launched the popular NCBI – BLAST application with its own 32-bit Windows Vista operating environment.
Both the Vista virtual environment of 6 GB per image and 1 GB of human genome data were provisioned with RepliCator parallel file serving to all cluster nodes simultaneously. A total of approximately 100 GB was provisioned in under 4 minutes to all nodes across Gigabit Ethernet. Provisioning with RepliCator occurs asynchronously, overlapped with other computation so processing never stalls for network I/O. With RepliCator, virtual provisioning is no longer inevitably associated with performance-degrading overhead. RepliCator removes a significant barrier to wider adoption of virtualization on clusters.
In the same tests, RepliCator's parallel file serving capability enabled the BLAST application and its Windows environment, still within the virtual container, to scale across the cluster with high efficiency. The virtually provisioned RepliCator-Vista application scaled linearly without degradation, while the same application run with NFS in the native Linux environment suffered significant performance degradation.
It is indeed time to make it so lethal that high I/O intensive apps, RAC, fat GUI apps, all will run happily on the hypervisor!
Comments
Post a Comment