Employing IP SANs for Microsoft exchange deployment.
Exchange is optimized for block storage. Most network attached storage (NAS) subsystems do not present themselves as block storage. While NAS can be configured to support Exchange, network file services impose greater processing overhead than block access storage such as iSCSI. For low cost and quick deployment, the majority of Exchange storage has been, and still is, directly attached. TO handle increased storage capacity needs and to service remote offices, many businesses and organizations simply install additional Exchange servers with local storage (see Figure 1). This creates significant management challenges and increased costs for today's budget-constrained IT organizations.
[FIGURE 1 OMITTED]
In an attempt to contain operating costs, many IT organizations are looking to server and storage consolidation, which can be achieved with storage area networks (SANs). Unfortunately, Fibre Channel-based storage area network (FC-SAN) solutions are often too expensive to justify and too complex to deploy.
The Internet Protocol-based IP SAN, utilizing the recently standardized iSCSI protocol, offers a more economically viable and cost-effective alternative to implementing shared, networked, block storage to facilitate Exchange deployments or server consolidations. By leveraging the learning from both the Internet development and the FC SAN deployments, IP SANs will quickly gain the storage resources (i.e., volume and devices virtualizations) and fault management intelligence to support easy storage expansion and re-provisioning for Exchange deployments with high availability requirements. Enabling Exchange Server Consolidation
The benefits of shared, networked storage are well understood. In addition to offering better storage utilization and scalability, networked storage can provide greater manageability and availability (i.e., supporting server failover by Microsoft Cluster Service). These are key attributes to realizing Exchange server consolidation projects.
Relative to low-cost DAS-configured Exchange servers, the high cost of FC SAN solutions (which have to be optimized for high-end transaction processing applications) often makes it difficult to justify Exchange server consolidation projects. The new IP SAN technology and products bridge that economic gap.
IP SANs transport SCSI commands and responses within the iSCSI protocol, which in turn are carried across the IP/Ethernet networks using standard TCP/IP protocols. This allows IP SANs to be deployed with low cost, familiar, local area network (LAN) infrastructure, as opposed to the unfamiliar, more complex FC network infrastructure. In addition, the TCP/IP connectivity inherently facilitates remote replications for disaster recovery without requiring FCIP routers/gateways.
As shown in Figure 2, Exchange servers connect to the IP SAN using either iSCSI host bus adaptors (HBAs) or an iSCSI driver on top of a TCP offload engine (TOE) or standard Ethernet network interface card (NIC). ISCSI HBAs and TOEs offload the protocol processing from the host CPU, while the iSCSI driver on top of the standard Ethernet NICs offers the low-cost, quick start alternative.
[FIGURE 2 OMITTED]
The shared storage of the IP SAN can be configured to suit different types of Exchange deployment supporting multiple types and scale of Exchange clients. To optimize storage access, Exchange log files and databases should be installed on different types of logical volumes. Exchange log files, normally accessed sequentially, can be placed on simple mirrored disks (i.e., RAID-1). Exchange databases, on the other hand, are primarily accessed randomly and benefit from striped mirrors (i.e., RAID-10).
Similarly, Exchange clients with different data availability needs may be assigned to logical volumes with different protection characteristics. For example, Exchange clients that rely on their e-mail for critical business needs can have their mail-store located on mirrored disks. On the other hand, casual e-mail users may have their mail-stores located on non-mirrored logical volumes and rely on the restore from backup tapes in case of database loss or corruptions.
With the improvement in reliability and performance, low-cost ATA disks are being deployed in the enterprise for fixed content and near-line applications. Performance benchmarks indicate that properly configured IP SANs with ATA arrays support Exchange deployment with no visible performance degradation to the Exchange clients. Such configuration has the added benefit of shared storage being cost effective to also serve the near-line, disk-disk backup of the Exchange log files and databases. The near-line backup enables quick recovery in case of Exchange database corruption.
Deploying High Availability Exchange Service
To protect from server failure and increase Exchanges availability, an active/passive Exchange server cluster can be deployed with the Microsoft Cluster Service. The Exchange Virtual Server can then be failed from the active to the passive node in case of a server hardware failure. The availability of the Exchange services is further enhanced with dual homing Ethernet LAN, as well as redundant or clustered IP SAN storage controllers. Figure 3 illustrates a high availability, clustered Exchange deployment.
[FIGURE 3 OMITTED]
With the clustered Exchange servers, when the primary Exchange server fails, the backup Exchange server connects to the volume set (i.e., the check-point file, the log files and the database) in the IP SAN and continues the service to the clients. In this case, the logical volume sharing between the clustered Exchange servers is controlled via the SCSI RESERVE and RELEASE commands.
To protect from network connectivity failures, the dual homed Ethernet LAN provides redundant paths between the Exchange server and the shared IP SAN storage. The links from a server are connected to two switches. The switches are interconnected to provide additional path redundancy.
In normal operation, one of the server links will be active. When the active link is disconnected or fails, the backup link assumes the traffic, albeit with a different IP address. In this case, the server will re-establish the iSCSI session to the IP SAN using the new IP address.
Alternatively, the redundant links of a server can be connected to a single, enterprise- class, high availability switch such as an aggregated trunk. In this case, the aggregated links share an IP address and the traffic load is distributed based on destination addresses in normal operation. When one of the aggregated links fail, all traffic is carried by the surviving link. However, the iSCSI sessions remain intact, as the link failover occurs below the IP layer and is transparent to the iSCSI layer.
In case of a switch failure, the LAN infrastructure will automatically reconfigure. Either the LAN bridging and/or the IP routing protocol will activate the alternative path. This network resiliency obviates the host-based path management required by the FC SAN.
On the IP SAN storage side, an iSCSI storage array with conventional, active-passive controllers or clustered, active-active storage controllers (such as the Intransa IP5000) presents one or more Ethernet links from each of the redundant controllers. These links are connected to the different switches, as is the case with server network connections. When the active link or the storage controller fails without warning, the iSCSI session will time-out and new session will be established to the back up storage controller.
With advanced IP SAN that utilize multiple, active-active storage controllers (as is the case of the IP5000) exchange databases and log files can be distributed on different logical volumes and the access to these volumes are handled by the different storage controllers in the cluster. This configuration further benefits from the load-balancing capability among the clustered storage controllers, where a storage controller can migrate some of volumes to peer storage controllers dynamically when under sustained, heavy I/O access. This dynamic volume migration is facilitated by the iSCSI session redirect feature.
The same dynamic volume migration capability of the IP5000 also enables transparent scheduled maintenance or upgrade. Specifically, a storage controller in an IP5000 cluster being taken out of service will issue iSCSI redirection requests to the Exchange servers, thereby transferring all of the iSCSI sessions from that storage to the remaining controllers in the storage controller cluster.
IP SAN For Advanced Exchange Deployments
Windows Server 2003 offers Virtual Shadow-copy Service IVSS). A VSS-enabled backup application(the VSS requestor) causes Exchange 2003 (the VSS writer) to quince its data volumes. The VSS then instructs the IP SAN with built-in virtual volume copy support, via the VSS provider agent of the IP SAN, to take a snapshot of the Exchange 2003 data volumes. Upon completion of the snapshot by the IP SAN, the VSS notifies Exchange 2003 to continue its normal operation and finally notifies the VSS-enabled backup application that the snapshot in the IP SAN is available for backup. This seamless snapshot operation minimizes the disruption to the Exchange service.
A key long-term benefit of an IP SAN is that of remote data access. As shown in Figure 4, IP SANs can be set up as inherent overlay networks on the enterprise IP backbone network. The IP SAN cloud can have multiple IP sub-nets and VLANs. Thus, both the "front-end network" between the application servers and the IP Storage servers, as well as the "back-end network" between the IP Storage servers and the IP Storage, reside within the cloud. Where high-speed metro-area network (MAN) is available, Exchange databases can be mirrored to a backup site across the MAN. To protect against regional disaster, snapshots of Exchange databases and log files can be replicated across the wide-area IP network. In case of disaster, the exchange servers at the remote site can be activated quickly using the mirrored or replicated data sets.
[FIGURE 4 OMITTED]
Finally, a number of other advanced IP SAN capabilities each as policy-based volume management, dynamic volume expansion, etc., will further simplify the Exchange service deployment and/or consolidation.
IP SANs leverage the broadly and commonly available IP network management knowledge and simplify IP SAN planning, deployment, expansion, and reconfiguration. The low-cost ATA disk-based IP SAN enables Exchange server consolidation while realizing the full benefits of shared networked storage. The fault-tolerant clustering and the intelligent volume management offered by the advanced IP SAN further enables the deployment of highly available Exchange services across the enterprise. All of these can be done within constrained IT budgets and human resources while establishing the foundation for easy expansion to continually meet the enterprise's growing messaging and collaboration needs.
Peter Wang is CTO of Intransa (San Jose, Calif.)
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Special SAN Section|
|Publication:||Computer Technology Review|
|Date:||Jul 1, 2003|
|Previous Article:||File systems and storage.|
|Next Article:||Roundtable: "switched on storage arrays": Part 2 of 3.|