Chinaunix首页 | 论坛 | 博客
  • 博客访问: 1020918
  • 博文数量: 74
  • 博客积分: 10000
  • 博客等级: 上将
  • 技术积分: 2950
  • 用 户 组: 普通用户
  • 注册时间: 2007-05-15 00:17






分类: 服务器与存储

2008-08-22 09:49:16

In addition to supporting failover services, Sun Cluster supports scalable services. Failover services
are deployed as a single instance (i.e., running copy) on only one node. Scalable services are deployed
using multiple instances – one instance running on each of multiple nodes.
HACMP also supports scalable services. CSPOC, for example, may be used to start instances of a
scalable application on multiple nodes with a single command. IP load balancing solutions may be
deployed with HACMP to distribute the load to these multiple instances.
HACMP is a more scalable solution compared with Sun Cluster 3.1. HACMP is built upon a set of
scalable services called RSCT (Reliable Scalable Cluster Technology). The RSCT design, derived
from the RS/6000 SP environment, enables formulation of large clusters that generate high volumes
of events. RSCT services are useable by any application.
Sun Clusters are not based on a robust, powerful scalable set of cluster services such as RSCT.
Therefore, it is unlikely that Sun Clusters will match HACMP clusters in terms of absolute cluster
size, scalability, and performance. As an illustration, HACMP has been supporting 32-node
configurations for some time. Sun Cluster 3.1 supports up to 16-node configurations. Moreover,
Sun Clusters support only one instance of a scalable service on each node.
HACMP supports up to eight nodes running RAC. RAC in a Sun Cluster requires a physical path to
all nodes while IBM does not have this limitation. Sun Cluster 3.1 supports only four-node RAC

Sun Clusters support a Global File Service. This cluster file system allows applications to be installed
once and run on multiple cluster nodes.
IBM offers the General Parallel File System (GPFS). The GPFS cluster file system allows up to 128
nodes to share a common file system view. Unlike Sun’s Global File Service, a GPFS file system may
be striped for higher performance (when transferring large amounts of sequential data).
With GPFS, applications may be installed once and run anywhere in the cluster. Furthermore, GPFS
is built on top of IBM’s Recoverable VSD (virtual shared disk) component, thus providing
transparent recovery of VSDs upon failure.

Sun Clusters support Global Devices.
Sun’s global devices enable their cluster file system (a proxy file system). IBM also supports a cluster
file system, GPFS. Global Devices also eases configuring a cluster by automatically discovering
storage devices. IBM offers the Quick Configuration Facility for easing the configuration of
HACMP clusters.

Sun Cluster 3.1 provides faster recovery.
Sun offers no proof for this claim and does not define what “faster recovery” includes. It may be
that Sun makes this claim based on continuous availability of core services such as the Global File
Service and Global Network Service. Therefore, when a failover occurs the time required to recover
these services is minimized. However, recovery of most Sun Cluster “failover” type applications
requires application restart – a significant contributor to overall recovery time. Sun Cluster
“scalable” applications differ as multiple instances are already running. In any case, HACMP
provides all that is needed for speedy recovery. Unless apples-to-apples comparisons are made, Sun’s
claim has no value.

Sun Clusters allow management of all cluster resources as if they were on a single system. Further,
resources are managed from anywhere on the network using a GUI.
HACMP provides a number of facilities that simplify management while they enable cluster and
systems management from a single system. CSPOC (Cluster Single Point of Control), for example,
provides a single point of service-, user-, and shared disk-management for all cluster nodes. SMIT
and VSD provide a cluster-wide GUI point for monitoring and managing resources. The Quick
Configuration facility enables fast cluster configuration via a drag-and-drop interface.

Sun Clusters support diskless failover minimizing recovery time. With the Global File Service,
applications need not reside on servers physically attached to their associated storage. Therefore,
application failover need not also require file system failover.
IBM supports a similar capability using RVSD and GPFS. GPFS requires each node to mount the
file system individually. This may be done automatically when the node is booted. With Sun’s Global
File Service, any mount of a CFS volume is seen cluster wide.

Sun provides monitoring agents for various applications. Further, the Sun Cluster 3.1 API allows
users to build custom agents to monitor applications.
IBM provides sample recovery scripts at no additional cost while Sun charges extra for most agents.
In addition, applications developed in conjunction with IBM’s ClusterProven program include
associated recovery scripts. HACMP 5.2.0 supports Application Monitoring configured through
SMIT. Using the Event Management capabilities of RSCT, Process Application monitoring detects
process failure. User-defined application monitoring checks the health of an application. In either
case, HACMP will initiate recovery by restarting or failing over applications. Unlike Sun Clusters,
HACMP does not require application-specific agents in order to monitor process failure.

Sun Cluster 3.1’s tight integration with Solaris 9 results in improved performance. Future integration
with Solaris 10 will improve matters even more.
Sun offers no proof of this claim. What does “improved performance” mean? In any case, in
failover clusters, the cluster software plays little or no role in application performance – only in the
speed of recovery. In RAC clusters, for example, the database and lock manager are the significant
determinants of application performance. Neither of the later two components is affected by Sun
Cluster 3.1’s integration with Solaris. In any case, HACMP is tightly integrated with IBM’s AIX
operating system.

Sun offers SNDR (Sun Network Data Replicator) for real-time data replication to a remote server.
IBM offers GeoRM for real-time remote data replication. Further, IBM offers HAGEO for failover
of applications between locations. HAGEO includes the functionality of GeoRM. Sun’s SNDR is
not integrated with Sun Cluster 3.1 so it cannot integrate remote data replication with failover.
HACMP V5.2.0 also offers a number of new disaster recovery features for extended distances that
Sun cannot match.

Sun Clusters supports agents to enable high availability support for a variety of common
IBM offers scripts and implementation guides to simplify implementing SAP and other applications
with HACMP. Such information is extensible to similar applications. Furthermore, IBM does not
charge for sample recovery scripts as does Sun for its application agents. Finally, the IBM
ClusterProven program has produced hundreds of applications tested on HACMP clusters.

Sun’s Enterprise Management Center enables monitoring of system elements and setting of
thresholds that, if reached, generate alarms. This includes adjusting heartbeats.
IBM offers industry-leading monitoring/alert capabilities via RSCT Resource Monitors. HACMP
includes numerous resource monitors that may be used to trigger various actions in response to
changes in thresholds. Customers and ISVs may develop application and resource-specific Resource
Monitors. These can work together to adjust heartbeats if necessary.

Sun Clusters enables dynamic addition of CPUs, memory, I/O, and disk arrays to clusters.
The Dynamic Reconfiguration feature of HACMP allows dynamic expansion of cluster resources.
Nodes may be added online. Further, individual nodes may be upgraded, with minimal disruption to
users on that node, using a rolling upgrade procedure.

HACMP is difficult to set up and administer and needs highly experienced administrators.
All clusters require experienced operators for optimal operating results. There is no evidence that
Sun Cluster 3.1 4/09 is better than IBM’s HACMP or any other cluster in this regard.

HACMP is not a modern product. Its antecedents and basic functionality derive from the older
RS/6000 and SP domains, which are old and obsolete.
HACMP has evolved over the years from its strong design base. Its functionality has increased to
meet the needs of new switches, cluster nodes, interconnects, file systems, and more. To the
contrary, rather than being old and oboslete, HACMP draws on its heritage to provide modern
features and functions based on a solid background. Sun cluster tecnology is so deficient that a third
party (VERITAS) sells far more cluster software on Sun hardware than Sun does.

阅读(1817) | 评论(0) | 转发(0) |