From: John Lee (thesunlover2002@yahoo.com)
Date: Wed Jul 17 2002 - 09:09:28 EDT
MANY people have been asking me the URL of the
document "VERITAS Cluster vs Sun Cluster". I have
contacted the sender already and will post the address
as soon as I get his answer. Sorry I can not send each
of you the pdf file simply because there are just too
many.
Here is the text file converted from the original pdf
file. It is readable but not that good. I don't know
if there is a better way to convert pdf to text. Let
me know if you know please.
Regards,
///////////////////////////////////////////////////
V E R I TA S W H I T E PA P E R
VERITAS Cluster Server
(tm)
2.0
vs. Sun Cluster 3.0.Table of Contents
Executive Summary . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .1
Significant Features and Limitations of Sun Cluster
3.0 . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .1
Global Devices . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .1
Global Network Service . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
.2
Global File Service . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .2
8-Node Support . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . .3
Storage Configuration Limitations . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .3
Quorum and I/O Fencing . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .3
Using Global Features . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .4
Scalable Services . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .4
Faster Failover . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . .5
Diskless Failover . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . .5
Other Features and Limitations . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. .6
Sun Cluster 3.0 API . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . .6
Supported Applications . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
.6
Notes on Supported Storage and Servers . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . .6
"It's in the Kernel" . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .7
Significant VERITAS Cluster Server Competitive
Advantages . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .7
Service Group Load Management . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .7
Service Group Dependencies . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . .8
Triggers . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . .9
Cluster Configuration and Deployment . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .9
System Administration and Management Tools . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .9
Wide Area Failover Support and Cluster Management . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .10
Summary . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . .10
Sun Cluster 3.0 Feature Comparison . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
.11.Executive Summary
The following document will detail the competitive
advantages of VERITAS Cluster Server(tm) 2.0 against
the Sun Cluster 3.0
product. The paper will highlight the significant
advantages of the VERITAS solution as well as note any
advantages in the
Sun Cluster product.
>From a high level, Sun is positioning Sun Cluster 3.0
as part of the Solaris vision, not as the endgame.
Overall, the campaign
is centered on "You don't need VERITAS anymore." Sun
is aggressively pushing new capabilities in Solstice
Disk Suite, UNIX
File System with Journaling and Sun Cluster 3.0 as the
beginning of a long-term strategy to place all
critical services in the
Solaris kernel. This is evident with Scott McNealy's
"one throat to choke" comments of late.
Sun Cluster 3.0 stacked up against VERITAS Cluster
Server 2.0 comes up short in many ways. However, Sun
is not likely to
engage in a feature-by-feature argument on a
per-product basis. Rather, it will quote the vision
and point on a lack of long-term
strategy from VERITAS.
The answer here is the total "business without
interruption" story. Complete availability management
in the local data
center and the wide area. Cluster Server, VERITAS
Global Cluster Manager(tm), VERITAS Cluster Server
Traffic Director and
VERITAS Volume Replicator(tm) provide capabilities
that Sun HA products may never attain. The key concept
here is that
"business without interruption" is the vision -
pervasive availability and availability management
across the enterprise, not
just the data center.
Significant Features and Limitations of Sun Cluster
3.0
Sun Cluster 3.0 embeds many features in the Solaris
kernel. Sun claims that this allows much tighter
integration of cluster
services between nodes. The private networks between
hosts, called the "cluster transports," are
extensively leveraged to
provide several features that are the key selling
points of Sun Cluster 3.0. These are Global Devices,
Global Network Service
and Global File Service.
Global Devices
Sun Cluster 3.0 uses Global Devices to provide
clusterwide access to any device in a cluster, from
any cluster node, without
regard to where the device is physically attached. A
global device name is created automatically when the
cluster software is
installed. By accessing the unique global device
identification instead of the local device name,
applications can access the
device from any node in the cluster. For example, a
CD-ROM on one node can be accessed transparently from
any other
node without first mounting and network file system
exporting the drive. Some devices, such as tape
drives, CD-ROM drives
and locally attached storage on a server are available
throughout the cluster, but access is lost if that
node fails. Dual-hosted
disk arrays can be made highly available and form the
basis for Global File Service.
The Global Devices concept goes beyond simple disks
and also works with logical volumes presented by
Solstice Disk Suite
and VERITAS Volume Manager(tm). Original device path
names are preserved and symbolic links point at the
global device. For
example, a VERITAS Volume Manager volume is usually
accessed via a device path such as
/dev/vx/dsk/oradg/oradata1. With
Sun Cluster 3.0, there is a symbolic link placed in
this directory that actually points to a global device
maintained by the
cluster.
Overall, the Global Devices concept is functional. It
provides a single view of storage hardware. A logical
volume manager
such as VERITAS Volume Manager or Solstice Disk Suite
already provides most of this functionality. The added
complexity of
Global Devices, coupled with the lack of performance
of the Global File System, make this "feature" hardly
worth the effort.
Page 1 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m.Global Network Service
Global Network Service provides a mechanism to have
multiple services running on multiple nodes, all
accessed through a
single IP. One host in the cluster actually hosts the
IP address (known as a global interface, or GIF) and
proxies network traffic
to the correct cluster host.
IP interfaces can be started or stopped dynamically
and migrated from one Solaris domain to another
without bringing down
the service. Finally, when a client accesses a virtual
IP, Sun Cluster can now load balance that IP request
across as many as
eight servers in a cluster.
This is another use of the cluster transport. IP
traffic arriving on the global interface is forwarded
over the transport to the
correct node. Sun classifies this as "scalable
services." The global interface node acts as a load
balancer for other nodes in
the cluster. Picture this as a built-in IP load
balancer, using the cluster transport to distribute
traffic to the correct node.
Load balancing is the one feature in Sun Cluster 3.0
that VERITAS Cluster Server does not provide today.
However, the load
balancing provided in Sun Cluster is extremely
limited.
Sun Cluster's load balancing:
Scales only to eight systems. This is extremely
limited compared to popular load-balancing products
such as Cisco Local
Director and Big IP.
Balances only IP addresses, ports and protocols - no
content (level-seven information) load balancing.
VERITAS Cluster Server's Traffic Director option,
announced in the second quarter of 2001, is based on
VERITAS Cluster
Server technology and will provide full-featured load
balancing. This product, scheduled to ship in the
third quarter of 2002,
will plug into the Cluster Server Cluster Manager as
well as Global Cluster Manager to provide a single
point of
administration.
We believe that because of the limited capabilities of
Sun Cluster's load balancing, it will not be a huge
selling feature over
VERITAS HA offerings.
Global File Service
Global File Service is a proxy-based file system that
allows up to eight systems to concurrently access the
same file system.
The architecture requires that each cluster file
system be mastered on one host, and all file system
operations flow through
this host. Global File Service has the master node for
accessing the disk, then uses the cluster transport to
service other
cluster nodes. A failure in a node with primary access
to the cluster file system is failed-over to a backup
or secondary node
and is transparent to user applications.
Failover is handled by client/server architecture. A
client on one node makes a request to a server on
another node. Each
server has a secondary node that can be used for
failover if necessary. The server sends "checkpoint"
messages (meta data)
to the secondary server for each write operation to
guarantee the integrity of data in case of failover.
Clients use the primary
as a caching system. This is implemented with a
caching scheme that makes access to data from remote
files almost as fast
as local access. The file system is implemented so it
can be used with a range of cluster interconnects. Due
to availability
issues with Sun Cluster Interconnect (SCI) hardware,
most Sun Cluster 3.0 systems will use 100BaseT or
Gigabit for cluster
transport.
In the case of the SCI-based Sun Cluster Channel, it
can take advantage of remote direct memory access
(DMA) to avoid
buffering information on the remote node and do a
direct memory access back to the requesting node.
However, this is a
Sun-only solution. All the hardware components
(interconnect, etc.) must be Sun-branded. So the
solution is very restrictive
and leaves the customer little room to maneuver.
Page 2 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m.The Global File System can be best thought of as
network file system over the cluster transport, with
tighter cache coherence.
On the surface, the Global File Service looks very
attractive. All file systems are accessible from any
cluster node. However, in
higher I/O situations, performance problems may
quickly come to light. The cluster transport is an
IP-based network, usually
running on 100 or 1,000 Mbit Ethernet. This adds the
latency of the IP stack into the picture. Because this
is a full proxy file
system, all reads and writes must be placed through
the master node. One can see that the master node's
kernel and IP
subsystem will rapidly become a bottleneck for I/O.
Heavy I/O load through the master will also greatly
impact its ability to
do compute loads in the cluster. As discussed later in
the "Scalable Services" section, Sun documentation
suggests using the
"HA Storage" agent to ensure a service with higher I/O
requirements is run on the same node that hosts the
storage, rather
than across the Global File Service.
VERITAS provides true concurrent access to data, and
thus continuous data availability, with the VERITAS
SANPoint
Foundation Suite
(tm)
HA solution set. The SANPoint Foundation Suite HA
bundles VERITAS Cluster Server with the new cluster
file system and clustered volume manager technologies.
VERITAS cluster file system technology provides a
significant
performance benefit over the Sun Global File System
approach because it provides direct access to the disk
via a fibre-channel-
based Storage Area Network (SAN), rather than sending
traffic over the production local area network (LAN).
One note of caution: although VERITAS supports up to
32 nodes in a traditional clustered environment, the
VERITAS
SANPoint Foundation Suite HA solution for concurrent
data access currently supports up to 16 nodes. Sun
states that its Sun
Cluster 3.0 with Global File Service solution will
support up to 8 nodes.
8-Node Support
Sun Cluster 3.0 currently supports up to 8 nodes in a
cluster. Customers can mix and match servers from
Sun's server line.
VERITAS Cluster Server continues to support up to 32
nodes in a cluster. In addition, while Sun is not
likely to support
anything other than Solaris clusters, Cluster Server
has a heterogeneous operating system roadmap. Cluster
Server supports
Solaris, HP-UX and Windows NT clusters. Windows 2000
and AIX support will be added in the first half of
2002. Linux
support will follow in the third quarter of 2002.
Storage Configuration Limitations
Sun Cluster 3.0, as currently shipping, has
significant limitations for possible storage
configurations. A maximum of 2 nodes
may connect to any external storage device. Each node
may have only one path to the storage. This prevents
use of any
High Availability storage access, such as VERITAS
Volume Manager dynamic multipathing (DMP). This
limitation precludes any
type of true N-to-N failover configurations.
Quorum and I/O Fencing
Sun Cluster 3.0 implements a rigid quorum scheme to
prevent split-brain conditions. Quorum requires a
minimum of 51
percent of cluster nodes be present to keep the
cluster in operation. To make this configuration more
usable, dual-hosted
disks can also be used as quorum devices. For example,
a 2-node cluster can add a shared disk as a third
quorum "vote" to
enable the cluster to remain operational after a
single node failure. During a cluster reconfiguration
after a node failure, the
remaining systems winning the quorum vote perform an
I/O fence operation to ensure data integrity. I/O
fencing blocks disk-write
traffic from nodes not present in the remaining
cluster.
Page 3 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m.Due to the requirement to work with third-party
storage vendors to implement SCSI-III persistent group
reservations (PGR),
Sun Cluster 3.0 currently supports only SCSI-II
reservation protocol. This issue severely limits
configuration options in the
cluster. SCSI-II reservation limitations preclude use
of any multiple paths to storage, such as VERITAS
Volume Manager DMP
or Solstice Disk Suite multiplex I/O. Further, the
maximum allowable hosts connected to any storage array
is two. So even in
an 8-node cluster, only two hosts may connect to any
storage.
VERITAS Cluster Server does not implement a quorum
protocol. In nearly all situations, the robust
heartbeat framework
provided by low latency transport (LLT)/global atomic
broadcast (GAB) is more than sufficient to provide
reliable cluster
membership services. This includes the capability to
place LLT/GAB heartbeat over the SAN fabric used to
access the disk
arrays (in specific tested configurations).
In reality, quorum protocols serve only one purpose.
They are designed to ensure a system that appears to
have left the
cluster is actually gone. A quorum protocol is
designed such that a system in the losing side of the
cluster will panic or
commit suicide. The issue is the time required for a
node to actually determine it should panic. During
this time, it still can
have I/O queued for disk. To guarantee no possibility
of a departed node writing to disk, a complete I/O
fencing protocol is
required. With a fencing protocol available, quorum is
no longer necessary.
To provide data protection in a full SAN fabric
configuration and address customer concerns about any
possibility of data
corruption, VERITAS is working with industry partners
to implement full SCSI-III persistent group
reservations I/O fencing in
the 2.0 Oracle Parallel Server release, as well as in
standard failover configurations in upcoming releases
of VERITAS Cluster
Server.
Using Global Features
The use of global networking, Global Devices and
Global File Service together are the core of several
features Sun positions
as superior.
Scalable Services
The Sun Cluster Scalable Data Service is a new type of
resource group that allows an application to online on
up to eight
systems in the cluster, and allows load to be
distributed across these systems. Service requests
come in through a single GIF
and are distributed to the nodes in the cluster based
on a load-balancing policy. Data storage is accessed
through the Global
File System. This would allow multiple copies of an
application to divide the customer workload and
provide horizontal
scaling. For example, a single IP could be used to
access a parallel application running across multiple
nodes. The GIF node
load balances requests, and each processing node
shares the overall processing load. Because all IP
traffic and data access
must happen over the cluster interconnect, one can see
where the cluster transport will rapidly become a
bottleneck.
Sun maintains that its scalability services enable an
application or series of applications to run
(distribute load) across multiple
systems in the cluster. While the concept of
horizontal scaling sounds very promising, actual
implementation is not quite so
simple. Most common applications were not designed to
function properly with multiple copies of the
application accessing
the same data. An administrator cannot simply start
multiple copies of an application and magically have a
scalable service.
Applications must essentially be designed from the
ground up to work in a parallel environment.
A good example is a standard Oracle database versus
the Oracle Parallel Server. A standard Oracle database
expects
complete ownership of all underlying data structures.
The database keeps in-memory copies of data present in
disk storage.
Because the database expects complete ownership, it
has no reason to check the on-disk copy for
consistency prior to
overwriting sections of disk storage. If multiple
Oracle instances were started with the same actual
database, each would
have no idea the other was present and overwrite
blocks of data. The Oracle Parallel Server is designed
to overcome this
problem by implementing a distributed lock manager to
keep multiple instances in synchronization with
regards to disk
storage.
Page 4 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m.Whenever scalability is discussed, Oracle Parallel
Service likely will be a key topic. Oracle Parallel
Service and scalable services
are two concepts. Oracle Parallel Service does not use
the Sun Cluster 3.0 Global File System or Global
Network Services. It
instead uses direct access to shared storage, and each
server node uses its own network connectivity. This is
a very important
detail. Oracle Parallel Service on Sun Cluster 3.0 is
in no way an endorsement of scalable services in Sun
Cluster 3.0. Oracle
Parallel Service uses Sun Cluster 3.0 for membership
services only. Oracle chose to completely bypass the
Global File System
to achieve the performance it needs.
VERITAS Cluster Server has provided parallel service
groups since version 1.0. With a proper shared data
infrastructure at the
back end, such as VERITAS SANPoint Foundation Suite or
Highly Available network file system via the VERITAS
ServPoint
(tm)
Network Attached Storage (NAS) Appliance, the customer
can implement parallel (scalable) services on up to 32
nodes. A
robust IP load-balancing solution such as Cisco Local
Director or VERITAS Traffic Director can handle
front-end, IP load
balancing.
Faster Failover
Sun claims that Sun Cluster 3.0 provides faster
failover through error detection, software switchover
and parallelized
application and infrastructure restarts. When a
failover occurs, clients might see a brief
interruption in service and might
need to reconnect after a failover has finished.
However, clients are not aware of the physical server
from which they are
provided the applications and data. While failover may
be faster in Sun Cluster 3.0 than it was in Sun
Cluster 2.2, there are
no studies that show Sun Cluster's failover is any
faster than a Cluster Server failover.
VERITAS Cluster Server already provides tunable
resource monitoring and parallelized restart of
nondependant resources.
Diskless Failover
Sun states that the foundation of the Global Devices,
Global File Service and Global Network Service
features embedded in
Sun Cluster 3.0 results in "diskless failover," which
provides greater flexibility and minimizes failover
time.
"Diskless failover" results from the ability of Global
Devices, Global File Service and Global Network
Service to create a
flexible, "virtual" set of data and application
resources that do not need to be tied to a particular
host or storage devices.
Data services do not have the requirement of residing
on dual-ported storage devices or being physically
attached to them.
Sun recommends using the Global File System for
applications with low I/O loading. For applications
with high I/O
requirements, it is recommended that the application
be configured with "storage affinity." This means the
node hosting the
application will also physically host the storage.
This is in itself a testament to the lack of
performance in the Global File
System. Diskless failover of an application with high
I/O requirements is therefore not recommended.
Diskless failover is an attempt to hide the underlying
fact that Sun Cluster can currently allow connection
of only two hosts
to external storage.
Page 5 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m.Other Features and Limitations
Sun Cluster 3.0 API
Sun provides a list of qualified High Availability and
scalable agents. If the application that a customer
wants to run as a High
Availability or scalable data service is not currently
offered by Sun, the Sun Cluster 3.0 application
program interface (API)
can be used to build support for an application to run
as a Highly Available or scalable data service.
With VERITAS Cluster Server, a customer can create a
custom agent to support applications or services not
provided as
bundled or enterprise agents.
Supported Applications
Based on marketing collateral from Sun, the
applications supported by Sun Cluster 3.0 are Apache
and Domain Name
Services, iPlanet, Netscape, Network File System,
Oracle and Oracle Parallel Server. These represent far
fewer applications
than what VERITAS Cluster Server supports. VERITAS
Cluster Server enterprise agents on the Solaris
platform are DB2,
Informix, NetBackup, Netscape, Network Appliance,
Oracle, Sybase and an upcoming iPlanet agent.
In addition to enterprise agents, there is a long list
of custom agents which VERITAS Consulting developed
(see
http://deptwebs.veritas.com/consultingservices/ecs%20offerings/ProLaunch/VCS%20ProLaunch/cust_agent.htm).
Customers
can create their own custom agents using tools
provided with VERITAS Cluster Server.
Notes on Supported Storage and Servers
A huge advantage of VERITAS Cluster Server over Sun
Cluster 2.2 was Sun's lack of support for any storage
that wasn't Sun
storage. This situation has changed a bit with Sun
Cluster 3.0:
Current support exists for D1000, A1000, A3500 and
A5x00
Sun has announced that Sun Cluster 3.0 will support
EMC storage in the future
Sun has announced that Sun Cluster 3.0 will support
the StorEdge arrays that are OEM'd from Hitachi Data
Systems
But it should be noted that Sun Cluster 3.0 was late
in supporting Sun's own T3 array. VERITAS Cluster
Server supported T3
(see the joint Sun/VERITAS press release) six months
before SunCluster 3.0 announced its support. VERITAS
Cluster Server
continues to have an advantage over Sun Cluster for
SAN and storage support.
VERITAS Cluster Server supports all clustering of
SunFire servers, including domain-to-domain failover
within and between
server boxes.
Page 6 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m."It's in the Kernel"
Sun claims that High Availability features must be in
the Solaris kernel to be "truly Highly Available."
This claim is misleading;
many features provided by the kernel do contribute to
a High Availability architecture. Examining the Sun
Cluster software
architecture in detail, one can see that the only
section of the entire architecture in the kernel not
present on VERITAS
Cluster Server is the global device access module and
"cluster networking" (global networking).
VERITAS Cluster Server places all node-to-node
communications and cluster membership in the Solaris
kernel with the
LLT/GAB modules. Cluster File System and Cluster
Volume Manager are also present in the kernel with
VERITAS File System
and VERITAS Volume Manager.
Both platforms place actual cluster operation in user
space, with the resource group manager (RGM) on Sun
Cluster and the
High Availability daemon (HAD) on VERITAS Cluster
Server.
With the weak functionality provided by the Sun
Cluster Global File System and lack of functions in
the global network vs.
true IP load balancers, these two features in the
kernel are not really an issue.
Significant VERITAS Cluster Server
(tm)
Competitive
Advantages
In addition to direct feature comparisons listed
above, the following section details areas in which
VERITAS Cluster Server
provides features not available in the Sun Cluster
product.
Service Group Workload Management
Sun Cluster 3.0 provides one possible method to
determine what node will act as a takeover node after
a failure. The order
of nodes listed in the configuration sets what node
will take over a group. The first node in the system
list in a running state
will be chosen. This very restrictive policy is
definitely not in line with Sun's claim that Sun
Cluster 3.0 will simplify data
center management.
VERITAS Cluster Server provides an extremely powerful
set of tools to intelligently and dynamically
determine a proper
failover host following an outage. VERITAS Cluster
Server provides a choice of three primary failover
policies: priority, round-robin
and load.
Priority is identical to the Sun Cluster 3.0 method.
The first available running system in the system list
is chosen. This is
ideal for a simple 2-node cluster, or a small cluster
with a very small number of service groups.
Round-robin chooses the system running the least
number of service groups as a failover target. This is
ideal for larger
clusters running a large number of service groups of
essentially the same server load characteristics (for
example, similar
databases or applications).
Load is the most flexible and powerful policy. It
provides the framework for true server consolidation
at the data center
level in large clusters. In large clusters, with tens
of nodes and potentially hundreds of applications, the
cluster software
must be able to decide the best possible system for
failover. Load policy is made of two components,
system limits and
group prerequisites.
System limits set a fixed capacity to servers and a
fixed demand for service groups. For example, a Sun
6500 is set to
a capacity of 400 and a Sun 4500 to 200. A database
may have a determined load of 150. On failover, the
server in
the cluster with the highest remaining capacity will
be chosen. When this group starts coming online, the
150 load is
subtracted from the server's remaining capacity.
Page 7 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m. The actual load on a server can also be
calculated in real-time by an outside application and
provide to VERITAS
Cluster Server through a simple command line
interface.
Service group prerequisites add capability to the
load policy. The user can set a list of finite
resources available on a
server, such as shared memory segments, semaphores and
others. Each service group is then assigned a set of
prerequisites. For example, a database may need three
shared memory segments and 10 semaphores. VERITAS
Cluster Server load policy will first determine a
subset of all systems that meet these criteria, then
choose the lowest
loaded system from this set. In this way, an unloaded
system that does not meet all the prerequisites of a
group will
not be chosen.
Overload warning provides the final piece of the
load policy. When a server sustains a predetermined
load level (static
or dynamically determined), the overload trigger is
initiated. (See the section titled "Triggers" for a
full description of
event management with triggers.) The overload trigger
is a user-defined script or application designed to
carry out
the proper actions. Sample scripts detail simple
operator warning about overload, as well as a method
to move or
shut down groups based on user-defined priority
values. For example, if load on a server running a
business-critical
database reaches and stays above a user-defined
threshold, operators will be notified immediately. The
overload
trigger then could scan the system for any service
groups with a lower priority than the database (such
as an internal
human resources application) and move the application
to a lesser loaded system - or even shut down the
application. The key here is that the framework is
completely flexible. The installer or user is free to
implement any
overload management scheme desired.
Service Group Dependencies
Like VERITAS Cluster Server, Sun Cluster 3.0 provides
the capability to link resource groups together to
control startup order
of different resource groups. For example, starting a
database group before starting an application group.
There are
significant limitations to the capabilities in the Sun
Cluster 3.0 product. Linked resource groups are
controlled only at start
time and then only if both groups are started on the
same node. Failure of a group will not cause failover
of the linked
group. Starting the group on different nodes, due to
operator error or misconfigured node lists, will also
cause the
relationship to break.
VERITAS Cluster Server has a much more comprehensive
set of service group dependency possibilities. VERITAS
Cluster Server
provides three possible online groups and one offline
group: online local, online global, online remote and
offline local.
In an online group dependency, the parent group must
wait for the child group to be brought online before
it can
start. For example, to configure a database
application and a database service as two separate
groups, you would
specify the database application as the parent and the
database service as the child. If the child faults,
the parent is
stopped and restarted after the child restarts. The
online group dependency has three forms:
In an online local dependency, an instance of the
parent group depends on an instance of child group
being online
on the same system. This is used typically in a
database and application service configuration where
the application
directly connects to the database.
In an online global dependency, an instance of
parent group depends on an instance of the child group
being online
on any system. This is typically used in a database
environment with a front-end Web server connecting via
IP.
In an online remote dependency, an instance of
parent group depends on an instance of the child group
being
online on any system other than the system on which
the parent is online. This configuration is useful
where the
load of the combined resource groups is too great for
a single system.
In an offline local group dependency, the parent
group can be started only if the child group is
offline on the system
and vice versa. This prevents conflicting applications
from running on the same system. For example, you can
configure
a production application on one system and a test
application on another. If the production application
fails, the test
application will be offline before the production
application starts.
Page 8 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m.Triggers
VERITAS Cluster Server provides a method for the
administrator to carry out specific actions when
events occur in the cluster,
such as a resource or group fault, or to carry out
specific tasks outside the actual service group before
it comes online or
goes offline. This capability is known as triggers.
Triggers provide the capability to extend the
functionality of the cluster
engine with simple shell scripts, perl or compiled
programs. For example, the PreOnline trigger can be
used to carry out
specific actions such as reconfiguring SAN zoning
before bringing up a group. The PostOnline and
PostOffline triggers can be
used to control applications outside the cluster, such
as signaling for a Web server restart following a
database switchover.
Cluster Configuration and Deployment
Sun Cluster 3.0 requires a very strict procedure for
creating and setting up a new cluster. The entire
cluster must be
described to the engine with individual commands, or a
script written to generate the same commands. For
companies
wishing to deploy large numbers of clusters, this can
be very time-consuming.
VERITAS Cluster Server 2.0 uses a human-readable
configuration called main.cf to describe the cluster
to the engine. The file
can be created with any text editor in a well-known
format. VERITAS also provides an offline configuration
editor to
manipulate a proposed or existing cluster
configuration from any Sun desktop with the cluster
manager installed. When
installed on a new cluster, the engine uses the
configuration file to build a new configuration and
distributes copies of the
file to all nodes. In practice, the format and human
readability of the file is a major time-saver. VERITAS
Enterprise Consulting
staff typically begins building a proposed
configuration while server hardware is being
installed. Loading the configuration
file into the offline configuration editor then allows
viewing complex resource dependencies in an
easy-to-use, Java-based
graphical user interface (GUI).
System Administration and Management Tools
Sun claims that the Sun Cluster 3.0 framework allows
all the cluster resources to be managed and
administered as if they
were on a single system. Administrators have access to
all system management tools from any system in the
cluster. Because
of the integration with the Solaris operating
environment, familiar commands execute as if only a
single system were being
administered. Cluster management is accomplished
through either a command line interface (CLI) or a
GUI, based on the
Sun Management Center framework. The administrator can
manage any resource on a cluster from anywhere on the
network where the Sun Management Center framework is
available.
Sun's claim applies to storage managed through the
Global Devices framework. Actual system administration
commands do
not run on all systems, unless the user types these
commands on the Sun-supplied console application. The
console
application routes typed commands to all systems via
the terminal concentrator. Several public domain
applications are
available to provide the identical functionality in
VERITAS Cluster Server.
For storage administration, VERITAS Volume Manager
provides the capability to manage storage connected to
any host in
the cluster. Actual functionality provided by the Sun
Management Center framework is severely limited. Sun
does not plan a
native Sun Cluster 3.0 GUI until the second or third
quarter of 2002.
Cluster Server has offered a superior set of
management tools for some time. Cluster Server
provides both CLI and GUI
management options. In addition, the integration of
Cluster Server and VERITAS Global Cluster Manager
offers wide area
failover functionality not available from Sun Cluster
3.0.
Page 9 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m.Wide Area Failover Support and Cluster Management
Sun Cluster 3.0 has zero wide area, disaster-tolerant
capability. Each cluster is a standalone unit.
VERITAS Global Cluster Manager is a very powerful tool
to manage multiple clusters across multiple sites.
Global Cluster
Manager provides heterogeneous cluster management, so
an administrator can control availability in an entire
enterprise,
across multiple platforms, from a single console.
Global Cluster Manager provides "cross-cluster event
correlation." This
allows an event in one cluster to initiate an action
in another. For example, the restart of a database in
a Solaris cluster can
cause a restart of a Tuxedo application in an NT
cluster.
Global Cluster Manager also provides complete wide
area failover capability through the integration of
VERITAS Cluster
Server replication [VERITAS Volume Replicator,
symmetrix remote data facility (SRDF) and Hitachi Data
Systems now, Oracle
Advanced Replication planned] and VERITAS Cluster
Server. Please see the Global Cluster Manager white
paper at
http://prodman.veritas.com/products/prod-detail/global_cluster_mgr/docs/pdf/Clustermanager_WP.pdf.
Summary
Sun Cluster 3.0 is a significant improvement over Sun
Cluster 2.2, but it is not anything close to VERITAS
Cluster Server 2.0.
VERITAS provides a much more robust solution with
VERITAS Cluster Server, across multiple platforms.
VERITAS Cluster Server
is the centerpiece in a complete availability
management platform from VERITAS, which includes
application clustering,
replication and full disaster tolerance.
Page 10 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m.Page 11 VERITA S C l u s t e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m
Global Devices Automatic detection of storage
devices on system boot-up,
including disk drives, tape drives
and CD-ROMs.
A nice feature, but not worth the overall loss of
functionality. VERITAS Volume Manager(tm) already
provides complete disk abstraction.
Global File Service Allows multiple servers to access
the same pool of data in the
event of a fault.
Global File Service is essentially Network File
System over
the cluster interconnect. Direct access to the Storage
Area
Network (SAN) fabric is required for true high-speed
access.
VERITAS SANPoint Foundation Suite(tm) HA provides a
superior solution by allowing direct I/O access to the
SAN.
Feature Description VERITAS Position
Sun Cluster 3.0 Feature Comparison
Global Network
Service
Allows an IP service to reside
anywhere in a cluster; multiple IP
addresses can be used for a
single service; and IP requests
can be balanced across as many
as eight servers in the cluster.
Global Network Service is a very
low-end/low-functionality
IP load balancer. If the customer is truly interested
in load
balancing, look into hardware-based solutions or
VERITAS
Traffic Director.
Remember that most applications that would typically
run
in a cluster are not capable of being run in a
parallel
configuration. A complex cluster like this is a bit of
overkill for Web servers.
Diskless failover Global Devices, Global File
Service and Global Network
Service enable application
failover that does not require
storage device or file system
failover and therefore is faster.
Sun recommends any application with heavy I/O needs
to be
connected directly to storage to minimize the
performance
hit. What application in a modern-day data center do
you
not want to have adequate disk performance?
Diskless failover is a clever ploy to hide the fact
that Sun
Cluster 3.0 supports a maximum of two nodes connected
to storage. If you want to host an application on more
nodes, you need diskless failover.
VERITAS Cluster Server supports full SAN fabric. You
can
host an application on 32 possible nodes, with full
performance I/O to the disk.
Scalable services The Sun Cluster Scalable Data
Service is a resource group that
allows an application to online
on up to eight systems in the
cluster online.
Most applications cannot be started on more than one
node without threat of data corruption.
For applications that can scale, VERITAS Cluster
Server
offers parallel service groups and SANPoint Foundation
Suite/HA for high-performance, parallel file system
access.
Other load-balancing solutions (such as Cisco Local
Director or Big IP) provide much more scalable
solutions
and are complementary to VERITAS Cluster Server.
VERITAS Cluster Server Traffic Director will provide
a
software-based, full-function IP load-balancing
solution in
the third quarter of 2002..Page 12 VERITA S C l u s t
e r S e r v e r
(tm)
v s . S u n C l u s t e r 3 .0 w w w.v e r i t a s .c
o m
Feature Description VERITAS Position
Sun Cluster 3.0 Feature Comparison (Continued)
Failover services Failover functionality is provided
by all cluster packages,
including VERITAS Cluster Server.
VERITAS Cluster Server SAN attach capabilities mean
a
service group can be failed-over to multiple nodes,
all
with direct SAN access.
Allows multiple services to share
the cluster. Multiple service
instances ensure that a service is
maintained in the event of a
failure.
Node support Supports 2 to 32 nodes in a cluster.
Heterogeneous strategy: the ability to create
clusters on
Sun, HP, Windows, AIX and Linux platforms (Q3 2002).
Supports 2 to 8 nodes in a
cluster.
Choice of failover
node
Ordered list of nodes in configuration file.
Round-robin. Least number of groups online.
Load. Complete flexibility to determine best-suited
node
at time of failure. VERITAS Cluster Server 2.0
supports
further features such as system limits and
prerequisites.
Ordered list of nodes in
configuration file.
Triggers Triggers provided for nearly all cluster
events, such as
node failure, pre- and post-online of a group,
heartbeat
issues, etc.
User-configurable extensions to
cluster functionality. Used for
event management and
notification. Not supported on
Sun Cluster 3.0.
I/O fencing Already provided with VERITAS Cluster
Server.
SCSI-3 reservations are currently provided with
Database
Edition/Advanced Cluster for Oracle, support for
standard
VERITAS Cluster Server will appear in the third
quarter
of 2002.
Uses SCSI-2 reservation to
prevent split-brain.
System
administration and
management tools
VERITAS Cluster Server also provides a Web GUI, JAVA
GUI and CLI.
VERITAS Cluster Server integration with Global
Cluster
Manager provides superior wide area cluster
management and failover.
Command line interface (CLI)
and graphical user interface
(GUI) that enables management
of the cluster.
Application program
interface (API) and
custom agent
development
VERITAS Cluster Server provides an API.
VERITAS will launch an agent developer's program in
mid-2002
to make it even easier for customers to create
custom agents.
Enables custom agent
development.
Storage Support VERITAS Cluster Server supports the
entire Sun Enterprise
line.
VERITAS Cluster Server supports a wide range of
storage
and SAN devices, including Sun's own T3 storage array,
EMC, Hitachi, Compaq, IBM and many other storage
vendors.
Supported storage and servers:
Sun Enterprise 220, 250, 450,
3x00, 4x00, 5x00, 6x00 and
10000; Sun D1000, A1000,
A3500, A5x00; OEM'd HDS.V E R I TA S W H I T E PA P E
R
VERITAS Software Corporation
Corporate Headquarters
350 Ellis Street
Mountain View, CA 94043
650-527-8000 or 800-327-2232
For additional information about VERITAS
Software, its products, or the location of an
office near you, please call our corporate
headquarters or visit our Web site at
www.veritas.com or e-mail us at sales@veritas.com
Copyright (c) 2002 VERITAS, VERITAS SOFTWARE, the
VERITAS logo, and all other VERITAS product names and/
slogans are trademarks or registered trademarks of
VERITAS Software Corporation in the US and/or other
countries. Other product names and/ slogans mentioned
herein may be trademarks or registered trademarks of
their respective companies. February 2002.
90-20328-399
Yahoo! Autos - Get free new car price quotes
http://autos.yahoo.com
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:24:36 EDT