building highly available systems: the dell poweredge … · rently supports two-node clustering...
TRANSCRIPT
storage
DELL POWER SOLUTIONS | May 200770
G lobal commerce and the demand for 24/7 access
have made information availability vital to
enterprise success. A lack of data access during
maintenance, hardware or software failure, or any other
period of downtime can lead to decreased productivity,
customer dissatisfaction, and lost revenue. As high-
availability technology continues to advance, both the
cost and complexity associated with clustering for avail-
ability and failover have been reduced significantly,
making these configurations feasible for small businesses
as well as large enterprise-class data centers.
For data to be continuously available, IT systems must
maintain uninterrupted data access during both planned and
unplanned downtime. Host-based clustering, in conjunction
with high availability at the storage level, can greatly enhance
IT service levels, and represents a crucial step toward ensur-
ing data availability and minimizing downtime problems.
Allowing multiple servers (for example, a two-node cluster)
shared access to data on a single storage system helps pro-
vide application availability even if one of the servers
becomes unavailable.
The Dell PowerEdge Cluster SE600W solution and Dell
PowerVault MD3000 storage array are designed to allow
enterprises of all sizes to build highly available systems in
a cost-effective way. Deploying this solution can help main-
tain access to critical application data even following a hard-
ware or software failure.
Dell PowerEdge Cluster SE600W Dell PowerEdge clusters are based on Microsoft Cluster
Service software and designed to keep applications and
services available during any single failure within the cluster.
When Microsoft Cluster Service detects a failure, it automati-
cally moves cluster resources from the failed cluster node to
a healthy one and restarts the applications. PowerEdge clus-
ters support cluster-aware applications such as Microsoft
SQL Server™, Microsoft Exchange Server, and Oracle® Database
with Oracle Fail Safe software.
In addition to application-level availability, enterprises
should consider redundancy in the server-to-storage I/O
path. Because a failure of any component along this path
(such as a server, adapter, controller, cable, or disk drive)
jeopardizes system availability, the storage system’s ability
to maintain data access during such a failure is a key part of
a highly available storage deployment.
The PowerEdge Cluster SE600W is the latest addition to
the Dell family of feature-rich clustering solutions, and the
first based on Serial Attached SCSI (SAS) technology. It cur-
rently supports two-node clustering based on the compo-
nents shown in Figure 1. The supported PowerEdge servers
provide key high-availability features such as error-correcting
code memory; software or hardware RAID for the internal
drives; hot-swappable drives, power supplies, fans, and PCI
slots; optional dual host bus adapters (HBAs); and optional
redundant paths to the storage systems.
Related Categories:
Clustering
Dell PowerEdge servers
Dell PowerVault storage
Direct attach storage (DAS)
High availability (HA)
Microsoft WindowsServer 2003
Serial Attached SCSI (SAS)
Storage architecture
Visit www.dell.com/powersolutions
for the complete category index.
Building Highly Available Systems: The Dell PowerEdge Cluster SE600W and PowerVault MD3000The Dell™ PowerEdge™ Cluster SE600W is a Serial Attached SCSI (SAS)–based cluster solution comprising Dell PowerEdge servers, Microsoft® Windows® operating systems, and the Dell PowerVault™ MD3000 storage system. Deploying this solution enables enterprises to achieve highly available service levels in a cost-effective way.
By NaM NguyeN
DaNiel Moges
shaBaNa M.
Travis vigil
reprinted from Dell Power Solutions, May 2007. Copyright © 2007 Dell inc. all rights reserved.
71www.dell.com/powersolutions
The PowerEdge Cluster SE600W offers three
configurations:
• Entry-level configuration: This configuration
offers only one path from each server to the
storage (see Figure 2). If a component in one
path fails, the server loses access to the
storage; when this happens, Microsoft
Cluster Service moves resources from the
affected server to the other server and
restarts the applications.
• Redundant configuration with dual-port
HBAs: This configuration provides higher
availability than the entry-level configura-
tion, incorporating two ports on each SAS
5/E HBA to create two paths from each server
to the storage (see Figure 3). If an HBA fails,
Microsoft Cluster Service handles applica-
tion migration the same way it does in the
entry-level configuration. If another storage
component fails, I/Os are transparently
rerouted to the alternate path without affect-
ing applications or clients.
• Fully redundant configuration: This configu-
ration provides the highest availability of
the three configurations, incorporating two
SAS 5/E HBAs and two paths from each
server to the storage (see Figure 4).
Administrators can configure I/Os to use
any path for load balancing. If a storage
path fails, I/Os are transparently rerouted
to the alternate path without affecting
applications or clients.
Dell PowerVault MD3000The Dell PowerVault MD3000 storage array, with
two high-performance, active/active external
RAID controllers and a mirrored cache, can be a
key part of designing a cost-effective, highly
available cluster system. It is based on SAS
technology and offers high-availability features
such as hot spares; hot-swappable drives,
power supplies, cooling fans, and storage con-
trollers; redundant storage controllers; redun-
dant back-end paths with dual-port SAS drives;
and optional premium features such as snap-
shots and virtual disk copy.
To appreciate the potential value of cluster-
ing with PowerVault MD3000 storage, enter-
prises should first understand the host-based
RAID method of clustering. This method uses a
direct connection from a storage device to each
server, and the RAID controller itself resides on
a PCI RAID card installed in the server. In an
external RAID configuration, like that of the
PowerVault MD3000, the RAID controller resides
outside the server in the external storage
system itself.
Among the issues associated with host-
based RAID clustering, one of the most signifi-
cant is related to the cache and performance.
The cache is high-speed memory designed to
reduce the time required to read or write data.
In host-based RAID configurations, the cache
resides on the RAID adapter card located in the
server. To help ensure that in the event of a
server failure the unwritten cache data is writ-
ten to disk, vendors disable the use of the
cache on RAID adapters to maintain cache
coherency for the cluster. Unfortunately, doing
Building Highly Available Systems: The Dell PowerEdge Cluster SE600W and PowerVault MD3000
Figure 1. Supported components for the Dell PowerEdge Cluster SE600W
operating systems Microsoft Windows server 2003 and Windows server 2003 release 2 (r2) enterprise edition and enterprise x64 edition
servers Dell Poweredge models 1800, 1850, 2800, 2850, 6800, 6850, 1950, 2900, 2950, 2970, and 6950
storage Dell Powervault MD3000
Figure 3. Redundant Dell PowerEdge Cluster SE600W configuration with dual-port HBAs
Figure 2. Entry-level Dell PowerEdge Cluster SE600W configuration
Figure 4. Fully redundant Dell PowerEdge Cluster SE600W configuration
reprinted from Dell Power Solutions, May 2007. Copyright © 2007 Dell inc. all rights reserved.
LAN DellPowerEdge
server
DellPowerEdge
server
HBA HBA
Controller 1 Controller 2
Dell PowerVault MD3000
Disk drives
LAN DellPowerEdge
server
DellPowerEdge
server
HBA HBA
Controller 1 Controller 2
Dell PowerVault MD3000
Disk drives
LAN DellPowerEdge
server
DellPowerEdge
server
HBA 1 HBA 2 HBA 1 HBA 2
Controller 1 Controller 2
Dell PowerVault MD3000
Disk drives
storage
DELL POWER SOLUTIONS | May 200772
so can cause a significant degradation in clus-
ter performance.
When using external RAID, the cache
resides on the external controller within the
storage system, and cache coherency can be
maintained without disabling the cache func-
tionality. For this reason, external systems
are preferable for a clustered configuration.
Storing the cache separately from the server
helps eliminate the risks associated with a
server failure in a host-based RAID cluster.
As an external storage system, the PowerVault
MD3000 is ideal for highly available two-
node clustering applications on PowerEdge
servers when performance and access to data
are critical.
Cluster features and advantagesThe PowerEdge Cluster SE600W and PowerVault
MD3000 are designed to build upon the reli-
ability of parallel SCSI while addressing that
technology’s performance, reliability, and scal-
ability limitations. Figure 5 summarizes the
cluster features of the SAS-based PowerEdge
Cluster SE600W compared with the parallel
SCSI–based PowerEdge Cluster SE500W.
Performance The PowerVault MD3000 can provide better
performance than PowerVault 22xS storage
because of its increased data transfer rate over
the SAS link and its enabled write cache. While
the maximum data transfer rate for Ultra320
SCSI is 320 Mbps (and the complications of the
shared bus architecture make increased speeds
unlikely), SAS currently offers speeds of up to
3 Gbps, and the SCSI Trade Association plans
for it to eventually support up to 12 Gbps.
In a PowerEdge Cluster SE500W, the write
cache is disabled in the RAID controller to main-
tain cache consistency, because the RAID intel-
ligence is part of the controller residing on the
host. In a PowerEdge Cluster SE600W, the write
cache is enabled in the RAID controller, because
the RAID intelligence is part of the controller
residing on the external storage device. Each
storage controller has 512 MB of cache, which
helps significantly improve overall system
performance.
reliabilityWhen deployed in an environment using
parallel SCSI and a shared bus architecture, a
misbehaving device can bring down the entire
system. The PowerVault MD3000 offers a point-
to-point architecture in which such devices can
be easily isolated without affecting the entire
system. In addition, the fully redundant PowerEdge
Cluster SE600W configuration provides two paths
from the servers to the storage, and includes
dual-port SAS drives to allow data access from
either port from one of the two storage controllers.
As a result, if a component in a storage path fails,
the multipath software can automatically reroute
the I/Os to the alternate path without affecting
the applications or clients.
scalabilityParallel SCSI can support up to 16 devices on
the shared bus, while a SAS domain can sup-
port up to 16,384 devices without performance
degradation. The PowerEdge Cluster SE600W
can support up to 45 SAS hard drives by adding
PowerVault MD1000 disk expansion enclosures
to the PowerVault MD3000 using a daisy-chain
topology (see Figure 6).Figure 5. Feature comparison of Dell PowerEdge Cluster SE500W and PowerEdge Cluster SE600W
Dell Poweredge Cluster se500W Dell Poweredge Cluster se600W
HBas up to two Poweredge expandable raiD Controller (PerC) 4/DC or PerC 4e/DC adapters per node
up to two sas 5/e hBas per node
storage up to four Powervault 22xs enclosures one Powervault MD3000 enclosure connected to up to two Powervault MD1000 enclosures
Cables 68-pin shielded P-type sCsi cable for connection to the host, up to 12 meters long
Compact cable and connector, up to 6 meters long
Hard drives 13 sCsi hard drives per enclosure 15 sas hard drives per enclosure (with expansion to up to 45 drives by adding Powervault MD1000 enclosures)
redundant path to the storage system
No yes, including redundant storage controllers with failover capability, dual-port sas drives, and a multipath failover driver
raID raiD-1, raiD-5, raiD-10, and raiD-50 raiD-0, raiD-1, raiD-5, and raiD-10
Write cache Disabled in cluster mode enabled in cluster mode
storage manageability
Dell openManage storage services and Dell openManage array Manager, installed on a host directly connected to the storage
Dell Modular Disk storage Manager with advanced features such as snapshots and virtual disk copy, installed on a management station (either in-band through the sas link or out-of-band over the network)
reprinted from Dell Power Solutions, May 2007. Copyright © 2007 Dell inc. all rights reserved.
“The Poweredge Cluster se600W and Powervault MD3000 are designed to build upon the reliability of parallel sCsi while addressing that technology’s per-formance, reliability, and scalability limitations.”
73www.dell.com/powersolutions
Highly available cluster solutionThe Dell PowerEdge Cluster SE600W and
PowerVault MD3000 take advantage of SAS
technology to offer enhanced performance, reli-
ability, and scalability, and include additional
features not found in parallel SCSI–based
PowerEdge Cluster SE500W configurations. This
cost-effective clustering solution can benefit
enterprises of all sizes that require high avail-
ability for critical applications such as data-
bases, messaging systems, and file, print, and
Web servers.
Nam Nguyen is a senior consultant in the High-
Availability Cluster Development Group at Dell,
and the lead engineer for SAS, Internet SCSI
(iSCSI), and Fibre Channel Dell PowerEdge Cluster
products. His current interests include business
continuity, clustering, and storage technologies.
He has a B.S. and an M.S. in Electrical Engineering
from the University of Texas at Austin.
Daniel Moges is a systems engineer adviser in the
High-Availability Cluster Development Group at
Dell. His responsibilities include developing SCSI-,
iSCSI-, and Fibre Channel–based high-availability
clustering products; his current interests related to
high-availability clustering include enterprise stor-
age technologies and database systems. Daniel
has a B.S. in Electrical Engineering from the
University of Illinois at Urbana-Champaign.
Shabana M. is an engineering analyst in the High-
Availability Cluster Development Group at Dell.
Her interests include the development of SCSI-
and Fibre Channel–based clustering solutions and
application performance. She has a B.E. in
Computer Science and Engineering.
Travis Vigil is a senior product marketing consultant
for Dell PowerVault storage. He has a B.S. from
Stanford University and an M.B.A. from Northwestern
University’s Kellogg School of Management.
Figure 6. Fully redundant Dell PowerEdge Cluster SE600W configuration with two daisy-chained Dell PowerVault MD1000 enclosures
reprinted from Dell Power Solutions, May 2007. Copyright © 2007 Dell inc. all rights reserved.
LAN DellPowerEdge
server
DellPowerEdge
server
HBA 1 HBA 2 HBA 1 HBA 2
Controller 1 Controller 2
Dell PowerVaultMD3000
Disk drives
Dell PowerVaultMD1000
Dell PowerVaultMD1000