Hyper-V Storage Enhancements & What They Mean for Users – Part 2: File Services Enhancements

hyperv blog 2
By Lawrence Gavin, Head Geek, SolarWinds®, Virtualization & Storage Management

This is part two of a three part blog series.
• Also refer to Part 1: Overview
•  Part 3: Cost Comparison of SMBv3 vs SAN Expansion

As noted in the introductory article, Windows Server® 2012 brings significant improvements to file services. In this article, we’re going to talk a bit more in-depth about each of these new SMB 3.0 features that makes file sharing a viable storage methodology for virtualization.

Storage Spaces – allows a file share to span multiple drive shares and multiple servers. The best feature of Storage Spaces is the ability to mirror or stripe a file share across multiple servers. A Storage Space appears to the file server as a local physical drive (with a drive letter), rather than as a network share, as was the case with the older Distributed File System (DFS). If you’re not implementing file server clustering, this can be an economical implementation to get nominal (software-based) redundancy in file storage services.

Cluster Shared Volumes – allows the sharing of a disk or LUN defined in the cluster shared storage that can be accessed by multiple nodes of the cluster simultaneously. This capability significantly enhances VM performance and availability.

SMB Transparent Failover – completely hides the failover operation from the system or application using file services, while maintaining full state between the session before failover and after.

SMB Scale-Out – leverages Cluster Shared Volumes (CSV) to provide multiple active file server nodes in a cluster. In addition to contributing to zero-downtime failovers, it also provides scalable file storage services, similar to other services that have been available through Network Load Balancing (NLB) services.

SMB Direct (SMB over RDMA) – Remote Direct Memory Access is a technology that allows the network fabric to directly access the data to be transferred in-memory from the remote server, completely bypassing the overhead of the TCP/IP protocol stack. SMB Direct is the RDMA implementation for SMB 3.0, allowing disk read/writes to occur from memory-to-memory across the network.

SMB Multichannel – the ability to use multiple data channels for network transfers. Each channel has affinity to a processor core. Typical network interfaces provide only one channel of data transfer, engaging only one core (regardless of the actual number of cores in the system). SMB Multichannel allows the use of Receive Side Scaling (RSS) to scale up the number of channels on the network interface to match the number of available cores. The allocation, and de-allocation, of channels is fully automatic and dynamic.

Storage Migration – the ability to physically move the files associated with a Virtual Machine (VM) from one location to another, without having to take the VM offline. In previous versions of Windows Server, this required a failover clustering infrastructure with shared storage, specifically CSV. Now you can migrate VMs between any Windows Server 2012 system. This allows you to live-migrate a VM off of Direct Attached Storage (DAS) in a standalone Hyper-V server to shared storage in a cluster, or from the shared storage back to a DAS-based Hyper-V instance. You can also live-migrate a VM from one DAS-based standalone Hyper-V server to another DAS-based standalone Hyper-V server.

There are three possible file server configurations that can take advantage of these capabilities:

• Single-node File Server – You can serve up files for VMs across multiple independent Hyper-V hosts. If you’re not particularly concerned about fault-tolerance, this can be a simple and inexpensive way to overcome the limitations of a limited number of spindles available with internal DAS.
• Multi-node File Server – A typical implementation when using clustering. Hyper-V nodes are clustered as compute nodes, and files are stored on a File Server Cluster using shared storage, typically Serially Attached SCSI (SAS), but could also be an iSCSI connection.
• Multi-node File Server with SAN – One of the expensive considerations with using a SAN for virtual hosts is that each host needs to have an HBA in order to communicate with the SAN. With these advances in file services technology, it is now possible to use Windows Server 2012 file services as a front-end to your SAN, eliminating the need to purchase, install, and configure HBAs in the Hyper-V hosts, as well as the necessary Fibre Channel switching infrastructure.

So these are the new technologies… But, you might ask, how well do they actually work? Here are two real-world examples of the performance capabilities of SMB 3.0.

• Mitch Tulloch describes an experiment in his book, Introducing Windows Server® 2012 (available for free from Microsoft TechNet), that streamed two video feeds from a VM, one a simple AVI file, the other a HD video file using a server-side transcoder. The VM was running at 30% CPU utilization. The migration occurred without dropping a single video frame from either stream.
• At MMS2013, Jose Barreto provided some advanced performance data in his presentation, File Storage Strategies for Private Cloud. In a live demonstration, he showed 16GBytes/sec of throughput using a single file-server with 6 SAS 6Gb/sec JBOD banks populated with 48 SSDs and a single Hyper-V host running twelve virtual machines connected with 3 RDMA-enabled 54Gb/sec Infiniband NICs.


Written by , Posted .

Leave a Comment

You must be logged in to post a comment.