EMC's VFCache Now Supports vMotion and Deduplication
August 28, 2012
SAN FRANCISCO--Server-side storage caching can drastically improve I/O performance by keeping frequently used data on SSDs locally on servers. However, when the caching is host-based, as with EMC's VFCache, VMware's vMotion won't work. EMC's VFCache 1.5, announced this week at VMworld, enables vMotion for hosts using VFCache and adds data deduplication, which squeezes more capacity from the cache, among other necessary enhancements.
VFCache is a server-side caching product using flash-based SSDs with a newly expanded 700-Gbyte capacity. VFCache runs within each VM and offers write-through caching, which means data is written to both the storage array and SSD simultaneously, but only the array confirms the write. Write-through caching doesn't improve write performance, but the data is better protected because it's verified to be stored on the array. The other option is write-back caching, which writes data to the SSD and then to the array. Write-back improves write performance compared with write-through, but there is risk of data loss if the cache fails.
- Big Customer Data Analytics with a 360-Degree View
- Datacenter Modernization: How Customers are Standardizing in Preparation for the Future
- Forrester Study: The Total Economic Impact of VMware View
- Five reasons why the right BI is like the right partner
VFCache runs in the virtual machine and the cache appears as a physical local disk, which hindered vMotions prior to vSphere 5.1. VMware's vSphere 5.1, also announced this week at VMworld, added support for live migration of VMs that have local disks and does not use shared storage. Even with the live migration of VMs on local disks like VFCache, migrating gigabytes of cache will take some time and slow the migration.
VFCache has a plug-in for vCenter that's used to initiate a vMotion rather than using VMware's manual vMotion option. Once a vMotion is started with the VFCache plug-in, the local cache is invalidated on the SSD so that data corruption doesn't occur; the VM is moved to the destination; and if there is a VFCache card on the target, a new cache is created and starts the warming (prepopulating) process.
Howard Marks, chief scientist at DeepStorage.net and a Network Computing contributor, points out, "Since VFCache requires extra steps carried out by EMC's software to migrate a virtual machine, it doesn't support automated vMotions carried out by VMware's Distributed Resource Scheduler. DRS creates and moves VM workloads across servers automatically. Vendors that install their caching components in the hypervisor--like Proximal Data and SanDisk's new FlashSoft for VMware--rather than in the guest OS can support vMotion and DRS without needing special user interfaces."
Breaking VMware's DRS will be a show stopper for organizations that use it to automatically manage virtual resources. Barry Ader, senior director of product management for EMC's flash business unit, says EMC is working on making VFCache work seamlessly with VMware's automated functions, which should be available in a future release.
VFCache also performs deduplication within the SSD cache, which conserves space by storing unique data once. There's a slight CPU and RAM hit to use deduplication since it's done in software, but Ader says the impact is minimal.
EMC also enhanced the hardware, adding support for as many VFCache cards as the server can accept. VFCache adds support for Cisco UCS LSI Nytro PCIe mezzanine cards available for the UCS B series computing blades. The LSI SSD cards are available from Cisco in 400- and 800-Gbyte capacities.
VFCache 1.5 is available now, and the LSI PCIe cards from LSI will be available in the fourth quarter.