You are on page 1of 16

Fundamentals of a Well-Built SAN

White paper

Introduction......................................................................................................................................... 2
Executive summary............................................................................................................................... 2
Anatomy of a SAN .............................................................................................................................. 3
The advantages of iSCSI SANs ............................................................................................................. 3
Core features of HP LeftHand SANs ....................................................................................................... 3
True clustering ................................................................................................................................. 4
Storage virtualization ....................................................................................................................... 4
Reliability with network RAID ............................................................................................................. 4
Continuous data availability .............................................................................................................. 5
Scalable capacity and performance ................................................................................................... 5
Easing administration overhead ......................................................................................................... 5
Enhanced data services .................................................................................................................... 5
Based on HP server technology ............................................................................................................. 5
Accelerated technology refresh rate ................................................................................................... 5
Investment protection ........................................................................................................................ 6
Lower administration overhead .......................................................................................................... 6
HP Services ..................................................................................................................................... 6
Scalable performance .......................................................................................................................... 6
Scalability by design ........................................................................................................................ 6
Limitations of traditional architectures ................................................................................................. 7
Objective scalability measures........................................................................................................... 8
Flexible scalability models................................................................................................................. 9
Scaling with tiered storage................................................................................................................ 9
High availability .................................................................................................................................. 9
Synchronous replication and disaster recovery with multi-site SANs...................................................... 10
Reliability by design........................................................................................................................... 11
Building reliability from the ground up .............................................................................................. 12
Extreme performance ......................................................................................................................... 14
Performance through parallelism ...................................................................................................... 14
Increasing parallelism and performance with MPIO DSM ................................................................... 14
Increasing bandwidth dramatically................................................................................................... 15
Protecting investments ..................................................................................................................... 16
Conclusion........................................................................................................................................ 16
For more information.......................................................................................................................... 16
Introduction
Next to its people, data is a company’s most valuable asset. Businesses of every size rely heavily on
data that is at once increasingly complex and increasingly regulated. Government regulations in
countries around the world dictate how financial, customer, personnel, and health-related information
is stored, maintained, and retained.
Central data storage and management help organizations achieve compliance with this mountain of
requirements, and help to ensure that business-critical data will be available when users need it.
Storage area networks, or SANs, are common solutions for businesses seeking to simplify storage.
Unlike direct attach storage (DAS), which creates disparate, random islands of information, SANs
centralize data storage. Disaster recovery solutions—a necessity in today’s world—are easier to
implement on SANs, and the stored data is easier to manage than in a DAS environment.
A company that leverages server virtualization for high availability and disaster recovery has even
more reason to deploy a SAN—and not just any SAN. Virtualized environments need shared storage
to take full advantage of the shared infrastructure. When both server and storage are protected, the
solution is complete.
Although Fibre Channel SANs are an option for large enterprises, these storage solutions tend to be
expensive and limited in their flexibility. Compact and cost-effective, iSCSI SANs are a better fit for
many businesses. HP LeftHand iSCSI SAN solutions are built in a way that’s fundamentally different.
Built from the ground up to be flexible, scalable, and highly available, HP LeftHand P4000 SANs
deliver all of the features of enterprise storage and are an excellent fit for disaster recovery, business
continuity, and virtualized storage solutions.

Executive summary
Your business’s increasing reliance on data calls for a well-designed and well-built data storage
system. Agile, easy to deploy, and intuitive to manage, HP LeftHand P4000 SAN Solutions provide
all of the functions that organizations expect to see in a Fibre Channel SAN—at an affordable price
point that makes centralized storage an economical option even for small- and medium-sized
businesses. Because HP LeftHand SANs are built with a superior architecture, they are more scalable
and offer higher availability, more reliability, and higher performance than other iSCSI SAN
products.
HP LeftHand SANs are based on iSCSI technology—SCSI over standard Internet protocols (IP). This
allows companies to use the standard iSCSI drivers that accompany server operating systems to
access storage over standard IP networks.
HP LeftHand SANs use a process called true clustering. True clustering means that every storage
system in a cluster participates equally in sharing both the cluster’s workload and storage capacity.
The cluster manages itself. With true clustering, organizations can administer a single entity while
configuring virtual volumes and per-volume network RAID levels. They can also take snapshots, make
remote copies, scale the cluster, and even take storage systems down for upgrades or maintenance—
all without affecting data availability.
HP LeftHand SANs deliver the enterprise storage management features that companies expect of Fibre
Channel SANs. However, the similarity ends there. The purpose of this white paper is to describe the
benefits of a clustered architecture in comparison with traditional controller-based architectures, and
to discuss the technology that makes HP LeftHand SANs better by design.

2
Anatomy of a SAN
Every SAN is deployed using a combination of software, hardware, and services. In this respect,
HP LeftHand SANs are no different from any other SAN. What make them different are the
advantages they provide in all three of these areas. HP LeftHand SANs run powerful HP SAN/iQ®
Software, which enables them to manage a set of storage systems as a cluster. They operate on
enterprise-class, industry-standard, x86-architecture-based hardware equipped with advanced disk
drives, processing elements, caches, and controllers. Finally, HP provides support and service that
keeps customer data available and protects it from loss.

The advantages of iSCSI SANs


As they evaluate technologies, many organizations choose iSCSI SANs because of the complexity
and costs of traditional Fibre Channel (FC) SANs. Installing a traditional FC SAN means purchasing
storage hardware years ahead of the need. This costly, “buy high, sell low” proposition is amplified
by ever-plummeting storage costs. The cost of a pair of redundant FC host bus adapters (HBAs), which
can approach that of a small server, impedes the use of centralized storage for all servers.

Growing a Fibre Channel SAN that uses the traditional model of a hardware RAID controller
connected to multiple disk trays has its limits: as the number of disk trays increases, the controller itself
becomes a bottleneck, requiring the purchase of more complete storage systems. And as 2-Gb/s to
4-Gb/s FC throughput is eclipsed by off-the-shelf 10-Gb/s Ethernet (10GbE), the perception that a FC
fabric delivers better performance than iSCSI is fast becoming obsolete.

Core features of HP LeftHand SANs


HP LeftHand SANs aggregate the resources of a set of enterprise-class HP servers into a storage
cluster. The result is flexible, scalable, virtualized storage (see Figure 1). The benefits of this
technology are discussed in the following sections.

3
Figure 1: HP LeftHand P4000 SANs are built using enterprise-class HP servers. The physical blocks in the SAN—corresponding
to virtual volumes—are distributed evenly across the cluster.

True clustering
P4000 SANs provide true, n-way clustered storage, not a traditional two-way active/active or
active/standby configuration. True clustered storage means that a set of storage systems is managed
and scaled as a single entity, with all of the cluster’s resources available to respond to requests. As
the cluster is expanded, the resources available to handle requests increase as well—alleviating the
problem of a controller bottleneck, which is commonplace when traditional SANs are scaled.

Storage virtualization
True clustering creates a virtual pool of storage, spreading the storage for its virtual volumes evenly
across all storage systems in the cluster. Storage can be reserved at the time that the virtual volumes
are created—or, through thin provisioning, it can be allocated only as disk blocks are actually
needed.
HP LeftHand SANs virtualize every volume across all the storage systems in a cluster—not within each
storage system, as do traditional Fibre Channel SANs. The sharing of storage across the entire cluster
results helps increase performance and storage utilization; in addition, it helps decrease management
costs and complexities as clusters are scaled.

Reliability with network RAID


The reliability of HP LeftHand SANs begins with RAID storage on each storage system. It is then
enhanced with network RAID, which replicates each block across the storage cluster up to four times.

4
When a virtual volume is populated, its data blocks are striped and replicated across the cluster’s
storage systems so that the entire cluster participates in the storage of every virtual volume.

Continuous data availability


Network RAID contributes to high availability, helping to ensure that the loss of a single storage
system does not result in data loss. If a server fails to the point where it must be taken offline for
repair, network RAID keeps the single server failure from resulting in loss of data availability. When
the server is replaced, or repaired and brought back online, the SAN automatically brings the storage
node’s data blocks up to date with the rest of the cluster.
The more storage nodes there are in a cluster, the less the failure of a single storage system can affect
performance. A traditional active/active dual-controller SAN product can suffer a 50% performance
decrease in the event of a controller loss. With P4000 SANs, the failure of a storage node affects
performance by, at most, a percentage equal to the proportion of the cluster that the storage system
represents.

Scalable capacity and performance


As a cluster becomes full, storage capacity can be increased by adding storage nodes. True
clustering enables considerable scalability. As new storage nodes are configured, the cluster
automatically readjusts its block allocation so that the allocated storage and the workload are once
again distributed across the cluster. This procedure immediately scales the cluster’s performance by
bringing incremental network CPU, memory, cache, RAID controller, and disk resources to the cluster.

Easing administration overhead


All storage clusters are managed through a single, intuitive centralized management console (CMC).
The ability to treat a storage cluster as a single entity—rather than a set of discrete storage devices—
eases administration costs and changes the cost model from one in which storage must be planned
and purchased upfront to one in which it can be purchased as needed.

Enhanced data services


In traditional SANs, enhanced data services are typically add-on features. With HP LeftHand SANs,
enhanced data services are included at no additional charge. Synchronous replication keeps data
highly available within a cluster, even if that cluster is geographically separated between sites in a
LAN or MAN. Remote copy provides disaster recovery for multiple clusters and sites from a single
interface, and failover and failback are automated features. The unique allocate-on-write thin
provisioning provides the ability to create volumes without dedicating physical storage to them.
Snapshots are space-efficient because they are thinly provisioned.

Based on HP server technology


P4000 SANs use cost-effective, reliable, and high-performance HP server hardware. The resulting
storage cluster offers greater performance and reliability than many purpose-built storage devices.
Basing storage systems on enterprise-class server hardware from HP adds to the superiority of
HP LeftHand SANs in a number of ways.

Accelerated technology refresh rate


The fast server technology refresh rate brings the latest technology to HP LeftHand SANs quickly. In
the storage industry, products tend to be refreshed every 24 to 36 months, but the server industry
pushes technology into customers’ hands on more of a 9- to 12-month cycle. In other words, the

5
benefits of higher-performance, higher-capacity drives, and the latest networking technology are often
available with HP servers—and thus HP LeftHand SANs—long before they are available with
traditional SANs.

Investment protection
Using HP servers offers unprecedented investment protection for organizations, because they can
build their entire server and storage infrastructure from the same basic components. They can
repurpose existing storage servers as application servers and, using the HP LeftHand P4000 Virtual
SAN Appliance (HP P4000 VSA™) Software, provide SAN storage on them as well.

Lower administration overhead


Leveraging HP server technology means having a more homogeneous environment to support, which
leads to greater staff efficiency. Purchasing HP LeftHand SANs often brings more HP technology into
the many data centers already using HP servers. This allows organizations to build upon their existing
infrastructure and knowledge, rather than investing in an entirely new, parallel infrastructure for
storage—with all of the associated training and management costs.

HP Services
HP LeftHand SANs are supported by global, world-class enterprise-support services. This can be a
significant advantage for data centers already engaged with HP. Companies may already have
spares on site; they may already know their support technician, and the technician may already be
badged for the site and familiar with its best practices. The size and skill set of the available
hardware support from HP also helps organizations compete.

Scalable performance
Scalability means that adding more resources to a system results in a commensurate increase in the
system’s ability to perform work. HP LeftHand SANs use true clustering to deliver a SAN that scales
both storage capacity and performance in a linear manner. Scaling a cluster with additional storage
systems supports growth by scaling existing volumes, adding new volumes, and supporting more
servers (Figure 2).

Scalability by design
HP LeftHand SANs are designed to deliver massive scalability. Unlike other clustered storage
products, P4000 SANs have no built-in limit on the number of storage systems per cluster. Because
true clustering allows them to scale performance and capacity linearly, HP LeftHand SANs differ from
traditional Fibre Channel SAN products.
HP LeftHand SANs are based on HP ProLiant servers, each of which has its own disk drives, RAID
controller, cache, memory, CPU, and networking resources. Thus, each time a new storage system is
added to a cluster, the cluster’s processing capacity increases in lock step with its storage capacity.
The SAN’s linear scalability derives from the fact that the ratio of processing resources to disk storage
is constant.

6
Figure 2: HP LeftHand SANs use true clustering to deliver linear scalability that can support more storage capacity and
performance as the cluster grows.

Limitations of traditional architectures


Compare this scalability model to the traditional controller/disk tray architecture that is behind most
Fibre Channel SANs and some NAS appliances (Figure 3). These systems use either an active/active
or an active/standby pair of processors, and scalability is achieved by adding more disk trays to the
configuration as more capacity is needed. Using this model, the ratio of processing power to disk
capacity decreases each time a new disk tray is added to the configuration. This causes performance
to increase to the point where the controllers and the Fibre Channel interconnect become a
bottleneck; performance then levels off or even declines. At that point, customers must either upgrade
to higher-power controllers or add new storage systems. Either action can result in significant amounts
of downtime.

7
Figure 3: A traditional SAN architecture scales by adding more disk trays to the configuration.

Objective scalability measures


ESG Lab, part of the Enterprise Strategy Group, measured scalability of HP LeftHand SANs of varying
sizes, and summarized the results in a July 2007 report 1 . The test cluster consisted of an HP LeftHand
SAN configured with up to 30 HP ProLiant DL320 Servers. Each server was configured with 12
300GB, 15k rpm serial-attached SCSI (SAS) drives, yielding a total of 3.6TB of storage per server.
ESG used the IOMeter benchmark to drive a simulated online transaction processing workload
consisting of 60% read and 40% write operations, using 8KB blocks. Scalability was nearly linear
from five to 30 storage systems, and the 30-server, 108TB cluster was able to sustain almost
50,000 input/output operations per second (IOPS).
HP LeftHand SANs allow organizations to build clusters whose characteristics are tuned for the
applications they support, scaling each cluster as needed. Many organizations deploy their first
open iSCSI SAN to support specific applications such as Microsoft® Exchange Server, VMware
Infrastructure 3, Microsoft SQL Server, and network file sharing. As they begin to see the benefits of
HP LeftHand technology first hand, they scale their cluster to support more and more applications
(Figure 4). As these applications become more diverse in their storage requirements, it often makes
sense to create a new cluster with a different type of storage system as its basis. Both clusters can
scale independently, and logical volumes can be moved between clusters with no interruption in
service. This makes scaling with tiered storage easy and straightforward.

1
“ LeftHand Networks 100 TB Enterprise SAN.”

8
Figure 4: Network RAID level 1 stripes and replicates a logical volume’s data across the cluster.

Flexible scalability models


HP LeftHand SANs scale capacity as well as performance. Organizations can increase the storage
capacity of a cluster by adding new storage systems, and they can even scale clusters with non-
identical storage systems. Scaling a cluster to 100 storage systems is as simple as scaling to two. In
addition to scaling the size of a cluster, companies can create multiple clusters because the cluster is
managed as a single unit.
Multiple SANs are configured in a single management group by the centralized management console
(CMC), which provides a single, intuitive graphical user interface (GUI) to manage them. Logical
volumes can be copied between clusters with a simple click of the mouse. Given the fact that
management groups can span geographical distances, remote copy works as simply and easily as a
local copy does between clusters.

Scaling with tiered storage


Different types of disk drives have different performance characteristics, and these characteristics
affect storage system performance and thus overall SAN performance. SAS and SCSI drives tend to
have high rotational speeds and low seek times, resulting in low I/O latency and therefore high
input/output operations per second (IOPS). Storage systems using SAS drives excel in supporting
transactional applications such as databases and Microsoft Exchange servers. Serial ATA (SATA)
drives tend to have lower rotational speeds and higher latency, but much higher capacity for the price
than SAS and SCSI drives. Clusters built around these storage systems are more cost-effective for
applications that demand high capacity, such as streaming media and general file storage.

High availability
Availability is about keeping data online and available at all times, an attribute that is typically
attempted by using redundant components. What happens if multiple components fail within a unit, or
the entire unit becomes unavailable?
The HP LeftHand P4000 SAN has redundant components such as redundant power supplies and
hardware RAID 5, 6, and 10. But, to achieve true high availability, volumes must remain online
whether a drive fails or the entire unit becomes unavailable. Network RAID provides this capability at
no extra charge.

9
Network RAID levels are assigned on a per-volume basis so that availability can be configured based
on the needs of individual applications and their data. This allows organizations to incur the cost of
redundant storage only for logical volumes that require it.
HP LeftHand SANs support striping plus network RAID replication levels 0, 2, 3, and 4, which
correspond to replicating each block up to four times. The most commonly configured network RAID
level, level 2, stripes and replicates blocks so that two copies of each block reside on the cluster
(Figure 4). Using this configuration, the logical volume continues to be available despite the failure of
a single storage node or the failure of two non-consecutive storage nodes. Compare this to products
that stripe without replication: the more servers in a cluster, the lower the availability—and the loss of
a single storage system means that all storage is lost.
In HP LeftHand SANs, clusters manage data layout and replication themselves so that failover is
automatic and so is failback. If a storage system fails and is later brought back online, the cluster
manages the process of restoring the repaired server’s data to the current level. Likewise, network
RAID level is a property of a logical volume that can be changed at any time. If the network RAID
level is changed, the cluster manages the process of increasing or reducing the replication level of
data for the logical volume as required.

Synchronous replication and disaster recovery with multi-site SANs


The availability properties of network RAID mean that functionality such as synchronous replication
and disaster recovery is built into the solution. This simplifies administration, helps to protect data, and
reduces costs compared to traditional SANs.
Volumes stored with network RAID level 2 have blocks striped across every other storage system, with
replicas stored across the rest. Logical volumes continue to be available even if half the systems in the
cluster fail.
Synchronous replication to provide multi-site high availability can be implemented simply by placing
half the storage systems in one location and the rest in another. Whether the alternate location is a
different closet in the same building, another building on the same campus, or a data center far
away, blocks are synchronously replicated across the two sites, and an entire site can fail without
making any data unavailable (Figure 5).
When the failed site comes back online, its storage systems automatically obtain any changed blocks
so that failback is automatic and transparent. With traditional SANs, this process can be time-
consuming and error-prone, and it can require application downtime.

10
Figure 5: Placing alternate storage systems in different, alternate locations enables synchronous replication and disaster
recovery as natural side effects of HP LeftHand SAN storage.

Reliability by design
Reliability is about protecting against data loss, which is a considerable proposition given the forward
march of disk technology. Disk reliability is typically expressed in terms of bit error rate (BER), which
means that disks may fail as a function of the amount of data read or written. As disk technology
allows drives to contain more data, they are more likely to fail when read from beginning to end; this
is exactly what happens during a RAID array rebuild after a single-disk failure.
The upward trend in disk sizes means that the dreaded second-disk failure that incapacitates a RAID 5
array is becoming more likely as disk sizes increase. Indeed, the chance of a second disk failure
while rebuilding a 9TB RAID array of SATA disk drives is nearly 10%—greater than the probability of
a dual controller failure in a traditional SAN. RAID 5 by itself is no longer sufficient. The reliability by
design of the HP LeftHand SAN allows organizations to configure logical volumes and clusters to
provide higher reliability levels.

11
Building reliability from the ground up
Reliability is built into every HP LeftHand SAN, from the choice of storage nodes to the additional
features that help organizations better manage their data.

• Hardware features—Every component in an HP LeftHand SAN includes dual power supplies,


network interface cards (NICs), environmental monitoring, and a battery-backed-up write cache.
• Choice of drives—The HP LeftHand SAN offers storage options that allow customers to choose the
combination of performance, capacity, and reliability that best suits their applications. Enterprise-
class SAS and SATA drives offer reliability levels equivalent to the Fibre Channel drives found in
traditional SANs, with a BER of 1/1016. Where capacity and price are more important than this
level of reliability, SATA drives can deliver a BER of 1/1014.
• Hardware and network RAID—Each storage system has built-in hardware RAID and battery-backed-
up write cache, both of which contribute to reliability and data protection. Depending on the system
and customer requirements, hardware RAID levels 5, 6, and10 can be configured on each storage
system and then combined with network RAID levels to maximize availability.
• Proactive self-healing—Just like traditional Fibre Channel SANs, HP LeftHand SANs monitor storage
systems for marginal environmental or drive conditions, allowing an HP LeftHand SAN to respond
with proactive support services, repairing potential faults before they result in an actual failure.
Each storage system is constantly scrubbing its storage to re-map bad blocks and restore data from
a storage system that has an intact replica.
• Geographic failover—Many organizations view the potential loss of data and application downtime
due to the failure of an entire data center as unacceptable. Network RAID allows for synchronous
replication and geographic failover so that business can continue without interruption. As discussed
in the section, “Synchronous replication and disaster recovery with multi-site SANs,” synchronous
replication is integrated into the superior architecture of HP LeftHand SANs. What’s more,
organizations can purchase an HP P4300 SAS Starter SAN for the price that some vendors charge
for their remote replication software alone. And, unlike the competition’s remote replication
software, failover and failback are not applicable to a HP LeftHand SAN because storage volumes
remain online and accessible throughout a site failure.
• Space-efficient snapshots—Every organization knows that there is no substitute for stable, offline
tape backups. HP LeftHand SANs support space-efficient snapshots where only the metadata—not
the blocks themselves—is copied to create the snapshot. Space efficiency means that the storage
system does not have to reserve an amount of storage for the snapshot equivalent to the size of the
volume itself. Space efficiency results in significant savings compared to traditional SANs.
Scheduled snapshots can be coordinated with application software so that the snapshot represents
a stable point-in-time copy of application data. In contrast to traditional SANs, snapshots in
HP LeftHand SANs can be mounted and written to by the backup software.
Snapshots may also be created manually and used to create temporary environments for developers
and administrators. Software upgrades and patches, for example, can be applied and tested to a
volume snapshot and tested on a real server or in a virtual machine before they are put into
production. Costs for these snapshots are low because an equal amount of additional storage does
not have to be available in order for the snapshot to be taken.
• Remote copy and asynchronous replication—The snapshot mechanism in an HP LeftHand SAN
forms the basis for remote copy and asynchronous replication capabilities. Snapshots contain an
identifier that ensures uniqueness within a management group. Snapshots can be copied between
clusters, promoted to be actual volumes, and then mounted and used. When clusters are
geographically separated, a local copy becomes a remote copy—there is no difference.

12
Consider an organization with data centers in Seattle and San Francisco, with New York acting as
a central data repository for backups or as a disaster recovery site (Figure 6). Snapshots can be
created in Seattle and San Francisco, copied to New York, and then promoted to be fully
populated volumes. Servers C and D, the users of the promoted snapshots, could represent hot
standby servers for disaster-recovery purposes, or they could be used to handle tape backup of the
copied volumes.

Figure 6: Remote copy is based on the snapshot mechanism, and it can be used for disaster-recovery and remote-archiving
purposes.

HP LeftHand SANs support scheduled snapshots and remote copies. Used in combination, these two
capabilities provide asynchronous replication, where batches of data representing the difference
between two snapshots are created and transferred between sites on a periodic basis.
Reliability is about protecting against data loss every step of the way, from establishing a per-server
reliability baseline to supporting geographic failover and failback capabilities. HP LeftHand SANs
provide flexible reliability at a price point that makes them superior to traditional SANs.

13
Extreme performance
The disk spindle’s performance characteristics ultimately drive disk storage system performance. One
challenge facing vendors of storage systems is how to extract the most performance from them.
In contrast with traditional SANs, HP LeftHand SANs excel at extracting top performance from storage
systems. Companies that use HP LeftHand SANs experience the high performance and capacity of the
latest disk drives—standard in the HP ProLiant servers on which LeftHand SANs are based—long
before those drives are available in traditional SAN architectures. In addition, HP LeftHand SANs
leverage the underlying hardware to extract an uncommon level of performance from the hardware
itself.

Performance through parallelism


Most SANs increase storage system performance through parallelism. HP LeftHand SANs allow
organizations to tune the level of parallel operations in their cluster to achieve required performance
levels.
Performance begins with RAID storage on each storage system, which uses parallel disk operations to
deliver high I/O bandwidth. Even though RAID 5 is the default configuration, higher and lower RAID
levels can be configured to adjust both performance and protection. For example, RAID 10 can be
configured for higher performance requirements, and RAID 6 can be configured to protect against
multiple drive failures.
A level above RAID on the individual storage systems, the cluster itself contributes to performance by
striping data—regardless of network RAID level—across the entire cluster. Just as RAID on the storage
systems delivers the performance of disk drives working in parallel, network RAID on the cluster itself
delivers the performance of multiple RAID arrays delivering data in parallel. Striping across multiple
storage systems is managed by the cluster itself, so administrators can manage only a single entity
rather than a set of individual storage systems.
Compared to traditional SANs, where the controllers themselves become a bottleneck as more disk
trays are added to a system, HP LeftHand SAN performance grows as storage capacity grows. This is
because each time a new set of disks are added to the cluster, they are supported by an additional
set of resources, including:
• CPU and main memory for managing storage, I/O, and the cluster
• Hardware RAID controller per storage system
• Additional battery-backed-up storage for each storage system
• Additional network interfaces that can operate in parallel

Increasing parallelism and performance with MPIO DSM


By default, an HP LeftHand SAN load-balances initial login requests from client iSCSI drivers across
the storage system, and then processes all subsequent requests from that client through a single
storage system. The storage system redirects each client request to the server owning the desired
block, and then redirects the response back to the client. Although this approach results in excellent
scalability, it is not as scalable as the performance that can be achieved with the HP LeftHand device-
specific module (DSM) for the Microsoft Windows® Multipath I/O (MPIO) iSCSI driver.

14
The MPIO DSM contains intelligence on the layout algorithms for the storage cluster. It can thus
calculate the location of any block in any virtual volume. Knowing which server contains the desired
block allows the iSCSI driver to contact the storage system that owns the block directly, without the
redirection used by the standard load-balancing approach. Figure 7 illustrates a redirected login
sequence, the SCSI mode sense command that loads the cluster-specific information into the driver,
and the separate I/O path that the driver establishes to each server in the cluster.

Figure 7: The HP LeftHand SAN’s device-specific module for the Microsoft MPIO driver (MPIO DSM) increases performance by
establishing parallel I/O paths—one to each storage system in the cluster.

The MPIO DSM provides the most benefit with sequential I/O. Performance scalability for standard
load-balancing is excellent. However, as the volume of data increases, the load that redirection
imposes on the network increases as well. MPIO DSM eliminates the additional data movement and
allows data to stream directly from storage systems to the client systems. Where the benefits of larger
clusters for standard load-balancing begin to diminish, the MPIO DSM configuration allows
performance to continue to climb significantly higher as storage systems are added to the cluster.

Increasing bandwidth dramatically


Most storage systems are limited in bandwidth. In traditional SANs, once storage capacity is
increased to the point where the system’s controllers can no longer provide increased bandwidth
along with increased capacity, customers either must upgrade controllers or purchase additional
storage systems. Multiple traditional storage systems add not only cost but also complexity. They also
need to be managed individually with volumes statically allocated to each system, resulting in
fragmentation. In contrast, a storage cluster can virtualize storage across the entire system.

15
Protecting investments
Investments in HP LeftHand SANs continue to be protected through the transition to 10-Gb/s Ethernet
(10GbE). Using standard, enterprise-class x86-architecture servers means using standard PCI Express
peripherals. Customers can upgrade existing clusters in the field with standard 10GbE interfaces.
Redundancy built into the underlying platform allows them to upgrade to a live cluster without
impacting data availability.
Unlike traditional SANs, each storage system in a cluster contributes to performance by delivering
data in parallel. So deploying a cluster with 10GbE provides improved throughput, which is then
enhanced further when that throughput is multiplied by the number of storage systems in the cluster.

Conclusion
Every SAN is built from a combination of software, hardware, and services, and in this respect iSCSI
SANs from HP are no different from any other SAN. But that’s where the similarity ends. HP LeftHand
SANs use distributed, clustered technology to deliver all of the functionality expected of a storage
area network. They add the advantage of linear scalability, high availability, per-logical-volume
configured reliability, and throughput of 10-Gb/s per node that is aggregated between all nodes in a
cluster. Better by design, HP LeftHand P4000 SANs deliver functionality including synchronous
replication, asynchronous replication, and remote copy. These are built-in features of a superior
architecture—not an add-on option that can cost as much as an entire SAN.
With true clustering, HP LeftHand SANs can virtualize storage across all storage systems in a cluster—
and even between clusters in the same management group. True clustering allows organizations to
treat a cluster as a single entity whose resources can be scaled and configured as needed. This helps
deliver the unique combination of performance and reliability that each application in a data center
requires.

For more information


For more information on HP LeftHand iSCSI SANs, visit www.hp.com/go/P4000.

Technology for better business outcomes


© Copyright 2009 Hewlett-Packard Development Company, L.P. The information
contained herein is subject to change without notice. The only warranties for HP
products and services are set forth in the express warranty statements
accompanying such products and services. Nothing herein should be construed as
constituting an additional warranty. HP shall not be liable for technical or editorial
errors or omissions contained herein.
Microsoft and Windows are U.S. registered trademarks of Microsoft Corporation.
4AA2-5616ENW, June 2009

You might also like