SAN Clarity in a Virtual Cloud
原文出處：Virtual Strategy Magazine
As the number of virtual servers sharing the physical FC HBA ports of a VOE server increase, SAN fan-out is no longer just a switch issue: SAN fan-out becomes a server HBA issue that mandates holistic management of SAN infrastructure from switches to HBAs.
Brocade HBA
For IT administrators, all of the popular HBAs, whether from Brocade, QLogic, or Emulex, provide a GUI-based management application: Brocade Host Connectivity Manager (HCM), QLogic SANsurfer, and Emulex HBAnywhere. Each of these applications has a very similar hierarchy of managed objects. That makes what each application can monitor and manage very similar. The critically important issue then becomes the ease and quickness with which IT administrators manipulate those managed-object hierarchies when performing important SAN maintenance and problem discovery tasks.
At issue is how well these tools empower IT administrators to quickly assess SAN traffic problems, accurately correlate problems with the performance of business applications, and correctly resolve processing disruptions. That's why SAN managers view HBA management within the context of holistic fabric management. How well HBA utilities manage their object hierarchies within a unified fabric context directly impacts productivity of IT administrators.
The importance of the impact that these HBA utilities have on system and storage administrators is underscored by an important IT heuristic: In the first year of operation, operating costs associated with managing storage hardware are often greater than the capital costs of acquiring that hardware. While many think of FC HBAs as commodity items, differences in the design and behavior of these HBA utilities will impact workflow and productivity for IT administrators.
To support business processes, IT must be able to quickly diagnose and resolve infrastructure events and issues that involve interdependencies among storage devices, hosts, and SAN switches. To simplify SAN connectivity, Brocade's Data Center Fabric (DCF) architecture provides an end-to-end strategic framework that strives to logically unify SAN HBAs, switches, and directors. The key component of DCF architecture, Brocade Data Center Fabric Manager (DCFM) Enterprise is the primary tool used by many SAN managers to assess data center fabrics-from storage ports to HBAs attached to either physical or virtual servers. DCFM's unified perspective provides IT administrators a significant advantage in maintaining, optimizing, and auditing a SAN fabric.
That unified perspective is all the more important given the explosive growth in the adoption of Virtual Operating Environments (VOEs). The limitations of software I/O profiles and the practice of isolating critical applications on dedicated servers left SAN fabrics with multiple servers for every storage device. Dubbed the fan-out ratio, with a high number of servers to storage devices, throughput at storage devices became a key bottleneck metric. Consolidation via server virtualization, however, radically changes that SAN topology.
With VOE servers hosting 8 or more virtual machines (VMs), an HBA in a VOE server can no longer be regarded as a simple commodity product. As the number of virtual servers sharing the physical FC HBA ports of a VOE server increase, SAN fan-out is no longer just a switch issue: SAN fan-out becomes a server HBA issue as well.
By extending the DCFM SAN management tool to HCM, the Brocade management tool for HBAs, Brocade also extends a well-established fabric management paradigm to the host side. As a result, IT administrators can apply Brocade's unique Quality of Service (QoS) Traffic Prioritization, which comes entirely out of the fabric, to each VM. Through unified host- and fabric-based Adaptive Networking services, Brocade HBAs provide support for policy-based data management and application service levels.
To assess the functionality and ease of use of server HBAs, VSM Labs focused on the ease with which an initial working SAN fabric can be configured and managed. Initial setup of new infrastructure is very important within a small- to medium-size enterprise (SME) environment as the impact of introducing new technology on a small IT staff is proportionally more disruptive. While the ease of initial setup remains important for new infrastructure, ease of setup pales in comparison to the question of how well any new infrastructure will help lower ongoing IT operating costs.
For the foundation of our infrastructure, we set up two Dell PowerEdge 1900 servers. Each server featured 4GB of RAM and a quad-core Intel Xeon E5335 processor. One server ran Microsoft Windows Server 2003, while the other ran Windows Server 2008 with Hyper-V, which in turn supported multiple Windows Server 2008 virtual machines. We installed three single-port 8 Gbit/sec HBAs in each server: A Brocade 815, a QLogic QLE2560, and an Emulex LPe12000. We then installed the GUI-based management application, which is independent of the HBA's throughput speed, for each HBA family: Brocade's Host Connectivity Manager (HCM), QLogic's SANsurfer, and Emulex's HBAnywhere.
To reflect the new edge-driven SAN fabric technology, and provide sufficient I/O throughput to meet the demands of our 8 Gbit/sec infrastructure, VSM Labs employed a Texas Memory Systems RamSan-400 solid state disk (SSD) array that was configured with four 4Gbit/sec FC controllers, each capable of handling 100,000 IOPS. We then provisioned one port on each controller with a target LUN. This topology resents a worse-case scenario for encountering potential SAN bottlenecks, as four 4-Gbit/sec data paths from the RamSan converge on each 8 Gbit/sec server HBA port.
The characteristically small IT staff at an SME site makes the installation and optimization of a SAN a daunting hurdle and the time taken by IT administrators to accomplished diagnostic and maintenance tasks all the more important. How quickly can a SAN fabric be up and working? How easy is it for system and storage administrators to keep a SAN functioning optimally? Will greater levels of resource abstraction lead to more administrator confusion? These questions shape the SAN evaluation criteria for any SME site.
To install each HBA, IT administrators begin by going to the vendor's web site in order to download the latest versions of all the required software, which includes drivers, firmware, boot code, management software and on-line documentation. It is here that the out-of-box experience for Brocade HBA users dramatically diverges from users of QLogic and Emulex HBAs. Installation should be a routine task that borders on the trivial. Only the Brocade HBAs, however, fully met our expectations-let alone surpass them.