Databases Reference
In-Depth Information
Table 4-1.
Workload Characteristics
Workload Characteristics
Constraints on Data
base
Fewer
More
Read-Write Mix
Low
High
Data Latency
High
Low
Consistency
Eventual
Immediate
Updatability
None
Constant
Data Types
Simple
Complex
Response Times
High
Low
Predictability
High
Low
Different workloads have different characteristics thus posing different challenges
when trying to support a mixed workload. Assuming the database is powerful enough to
support and is specialized for one particular type of workload is fine, but in the real world
there is always a mixed workload.
Read-Write Mix:
Whenever we use a database, the workloads are a mix of reads and
writes. However, between OLTP, BI and analytics needs you will see these mix of read
and writes taking different forms. OLTP is a write-intensive workload whereas BI and
analytics are thought of as read-only. Most BI systems write data in bulk at one time and
multiple read operations afterward whereas OLTP reads and writes happen at the same
time. The intensity of reading and writing and the mix of the two are important aspects of
a workload. Business intelligence-specific databases designed to handle read-intensive
work are often designed to load data in bulk. While the bulk loads are happening, it is
advised not to initiate any other write operations or queries.
Operational BI and dashboards often require up-to-date information. Analytic
processing is done in real time as part of the work in OLTP systems. The workload for an
operational BI application can look very similar to an OLTP application.
In case of big data scenarios, where many of analytic workloads are based on log data
or interaction data, you can expect a high volume of data flowing in continuously, so it
must be written continuously. Continuous loading is the extreme end of the spectrum for
write intensity. Likewise, in large-scale analytics, particularly when building analytical
models, entire data sets are read one or more times, making them among the most
read-intensive workloads.
Data Latency:
Data latency is the time lag between creation of data and usage of
data. Based on the business needs, applications can have different tolerances for latency.
For example, OLTP systems have short latencies, with the data available for usage as soon
as it has been inserted or updated; whereas data warehouses have long latencies, updated
once per day. Short latencies impose more restrictions on a system.
Longer latency requirements mean you have more flexibility in marshaling the
database resources. Respite from latency allows you to architect your data management
processes in a different way: incremental updates or bulk data processing in a batch
mode. The separation of data collection processes from data consumption processes
gives flexibility in designing data stores and puts fewer restrictions on a system.
