MarkLogic Server 11.0 Product Documentation

Scalability, Availability, and Failover Guide — Chapter 2

Terminology

It is important to understand the following terminology when considering a distributed implementation of MarkLogic Server:

The following diagram illustrates how the concepts of clusters, hosts and groups are implemented in an example multi-host distributed architecture. The diagram shows a single cluster involving six hosts that are segmented into two groups:

This second diagram shows how a single database called Documents could comprise four forests spread across the three hosts in the group Data-Nodes from the above diagram. The diagram also illustrates two internal databases maintained by the system for administrative usage:

Fundamentals

With the key terminology introduced, here is a short list of fundamentals to keep in mind while considering any multi-host deployment:

• All hosts in a cluster must run on the same platform.

  Because MarkLogic Server takes advantage of platform-specific optimizations, it is not possible to mix and match platforms in a single cluster. This means that you will need to consider the pros and cons of different CPU architectures and operating systems across all hosts in your cluster when planning a multi-host deployment.

• Backups are not portable across platforms.

  Not only must clusters be platform homogeneous, the internal data formats used by forests are platform-specific. Consequently, forests and forest backups are not portable across platforms. To move database or forest content from a MarkLogic Server implementation running on one platform to an implementation running on another platform, the database must be unloaded from the first system and loaded from scratch into the second database.

• All hosts in a cluster run identical software.

  Even though each host in a cluster can be configured to perform a different task, the full MarkLogic Server software runs on each host. Consequently, software installation is identical across the nodes. However, hosts can be reconfigured rapidly--on an individual or cluster-wide basis--to perform whatever role is required of them. Of course, it is quite possible that different hardware configurations will be used for different hosts in the cluster, depending on the roles they are expected to play.

• Front-end query distribution is the responsibility of external infrastructure.

  Multi-host deployments may incorporate multiple hosts running the same HTTP, WebDAV, or XDBC server applications, allowing queries to be distributed between those hosts for improved performance. MarkLogic Server transparently manages the distribution of content storage across forests within databases and the distribution of query execution across the hosts managing those forests. However, MarkLogic Server does not manage the distribution of queries across front-end hosts configured with equivalent HTTP, WebDAV, or XDBC server applications. The multi-host deployment relies on surrounding infrastructure for this load-balancing task.

• There is no master host in the cluster.

  Many distributed architectures rely on a single designated master to direct and manage the operations of the cluster. In MarkLogic's distributed architecture, all hosts are considered equal, regardless of the roles for which they are configured. Consequently, there is no designated master in the cluster. However, there are one or more bootstrap hosts in a cluster that handle the initial connection between clusters.

• Administration can be carried out through any host offering appropriate services.

  With no designated master in the cluster, administrative operations can be conducted on any cluster host configured to offer browser-based administrative operations or to accept XQuery-based administrative directives.

• Each forest is assigned to a single host.

  All queries that need to check for content stored in that forest or retrieve content from that forest will be routed to that host for processing. If you are using local-disk failover, you can replicate a forest across multiple hosts (for details, see High Availability of Data Nodes With Failover).

Advantages to Distributing

A single MarkLogic Server architecture can constrain its scalability and performance in several ways:

• Query capacity is constrained by the number of CPUs that can be put to work processing queries simultaneously.
• Content volume can be constrained both by available address space and by the amount of memory installed in the server.
• Data access rates across multiple queries can be gated by the limited I/O path(s) inherent in single server architectures.

The MarkLogic distributed architecture has been designed specifically to address these issues. Distributed deployments can be used to support scalability and performance across multiple dimensions:

• Large volumes of content can be supported by increasing the number of hosts hosting forests.
• High data access rates can be supported by distributing the storage I/O requirements across more hosts.
• Large query volumes can be supported by increasing the number of hosts performing query evaluation.

In addition, the MarkLogic distributed architecture enables significantly more cost-effective implementations. MarkLogic allows data centers to distribute workload across a number of less-expensive entry-level servers or bladed infrastructures rather than requiring an investment in large expensive SMP systems.

For data centers with existing investments in SMP architectures, MarkLogic can still make effective use of those platforms.

Considerations for a Distributed System

Just as there are different reasons to deploy a distributed architecture, there are different issues to consider in determining a distribution strategy. Each strategy provides certain benefits and involves certain other compromises. In order to determine how best to deploy MarkLogic Server to meet your requirements, you should obtain assistance from MarkLogic's professional services group.

This section includes the following topics:

• Hosts with Distinct Roles
• Scale of System
• Hardware Configuration
• Storage Architecture

Hosts with Distinct Roles

Distributed deployments of MarkLogic Server can be configured using hosts which all play similar roles or hosts which play distinct roles, as illustrated below.

A first-cut approach to small-scale distribution will have a number of hosts, each providing query evaluation and data service capability. This model can work well in static environments, and when the anticipated loads for each role map well.

In more dynamic environments, when the scale of the environment relative to individual host capacity is large, or when the anticipated scales of the query evaluation and data service loads differ markedly, the recommended approach is to partition the hosts into different roles. This architecture makes it easier to scale role-based capacity independently, and can also have some configuration advantages when different services operating at different scales are being offered out of the same cluster.

Configuring a Cluster

When first configuring and deploying a cluster, the best place to start is with the overall architecture and distribution strategy. With the right strategy in hand, you can rapidly develop the actual configuration parameter sets that you will need. This section includes the following topics:

• Building a Cluster from Scratch
• Increasing the Size of an Existing Cluster