IRJET- Improving Performance of Data Analytical Queries using In-Memory Database Systems

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 09 | Sep 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 295
Improving Performance of Data Analytical Queries Using In-Memory
Database Systems
Syed Ateeq Ahmed
Dept. of Computing, Middle East College, Oman
------------------------------------------------------------------------***-------------------------------------------------------------------------
Abstract - Improving performance of data analytical
queries needed a different approach than the online
transaction processing queries. Traditional databasesystems
which store data in row format in memory are more relevant
for transaction processing requirements. To achieve
significant performance gains of data analytical queries,
columnar format is more relevant. In-memory database
systems support storing data in columnar format and are
more relevant for improving performance of data analytic
queries. In this paper, architecture of in-memory database
systems is discussed and compared with traditional database
systems. This paper provides a solution for performance
enhancements of data analytical queries and the need for
using in-memory database systems in decision-making.
Key Words: Database Systems, In-memory database
(IMDB), Random access memory (RAM), Online transaction
processing systems (OLTP), Solid-state drives (SSD).
1. INTRODUCTION
Database management systems (DBMS) are software’s used
to create and work with databases. These database
systems are very popular and used in almost all
organizations where data storage, retrieval and processing
of data are one of the key requirements of the organization.
Main memory plays an important role in the performance of
the database. Database system caches the information in
memory and to preserve persistencychangesmust bestored
on disk. Compared to the memory access, reads from disk
take considerably more amount of time around in the order
of 10 milliseconds. Physical input and output also has an
impact on CPU resources. It is highly recommended that
repeatedly accessed objects if cached in memory will have a
positive performance impact..
1.1 Type of memory and its impact
One of the reasons for usage of type of database system is
based on which of type of storage is used to keep the data.
Though there are various types of storage available, they can
be classified in the three groups as solid-state drives (SSDs),
hard-disk drives (HDDs) andrandom accessmemory(RAM).
Each of these types of memories vary in access time, speed
and price.
Over the years,as thesizeofmemoryisgrowingcontinuously
and in contrast the price is falling drastically by a factor of 10
every 5 years [7]. The memory typesconsideredarerandom
access memory (RAM), solid-state drive and hard disk drive.
In case of in-memory databases, RAM is used to store copy of
data and solid-state or hard disk can be used for persistent
storage whereas traditional database systems uses RAM for
caches and persists data is stored on SSD or HDD. RAM is
faster than the SSD and HDD, whereas SSD is faster than the
HDD. In terms of price, HDD is cheapestamong the indicated
types of memories. SSD is around ten times more expensive
thanan HDD and RAM is ten times more expensive thanSDD.
If the required access time is less than 1ms, RAM is the
preferred choice. However, SSD can be an option if the
access time is up to 100 MS. If the access time is not critical,
and requirement is such that more than 100 ms is suffice,
storing data in in hard disk drive can be considered.
Type of access, sequential or random also plays a role in
choosing type of storage. Cost is directly proportional to the
storage speed, cost will be higher if the storage is slower.
Slow storage are useful for sequential access of data. If the
acceptancelevel of sequential access oflarge data volumesis
up to 100 MB/s, hard-disks will be an cost effective storage
option.
Data that should be accessed online with less than 1 ms, it
must be stored in the primary storage i.e. RAM. This kind of
data access needs to be stored in an in-memorydatabaseasit
is the most efficient solution for memory-resident data.
Comparatively, a relation database, is several times slower.
One of the main disadvantages of in-memory databases is
their relatively poor performances in the multiple CPU core
environments.
In any organization there will be data which need not be
accessed online and the requirement is such that data access
speed is acceptable up to 100 ms can be stored in a relational
database or an in-memory database depending on certain
factors. If the storage is not having any effect on the total
cost, it can be stored in an in-memory database or vice versa.
In-memory database systems sometimes are also used to
store offline data as they will utilize the CPU more efficiently
than the relational database systems. To avoid any further
workload on CPU and disks, unstructured data in the
organization for example, user profiles etc. can stored using
the regular file systems. Usage of file systems in such a
situations can help in improving the performance of the
database systems. The historical data for analytical purpose
can be stored in hard disks as the access type is mostly
sequential.
1.2 Pros and Cons of Choosing an in-memory
database system for an organization
The following factors must be consideredwhilechoosing the
in-memory database systems for any organization [5].

• In-memory database systems are used to improve the
performance of underlyingquerieswhichaccessdata in
the applications. Fewer activities and I/O operations
are required to access the data due to the reduced
instruction set.
• As the price of memory continue to decline and most of
the servers are having 32 TB or more of memory, to
achieve performance gains in-memory database
systems can be considered.
• Most of the organizations are providing real-time
interfaces and web-based applications, the need of
speed is the key requirement. In such situations, high-
end performance provided by the in-memory database
systems can be useful.
• The current in-memory processing technology,issueof
persistence is solved. The data in primary memory
persists even in case of power failures.
• With increased data reliability and persistence, in-
memory database systems are feasible for both
analytical and transactional processing needs of
organizations. Some of the examples can be
networking and telecom, capital markets, defence and
intelligence, gaming, real-time analytics etc.
• Part of the reason for not migrating to in-memory
database is cost. Although memory is getting cheaper
every year, it is still costlier than disk.
• Lack of in-memory database expertise and use of
interfaces other than pure SQL can be barrier in
migrating to in-memory database systems.
• Many still believe that the size of the in-memory
database must be limited in size, however, some of the
in-memory databasesystemssupportuptoterabytesof
data.
• Tests should be conducted to ensure that in-memory
database systems are a viable option to obtain
performance gains.
2. In-memory database systems and persistence
In case of in-memory database systems, complete dataset is
kept in the primary memory. User requests for data access
and updates to the existing data are done in the primary
memory without involving the relatively slow disks.
Persistence is a key requirement of a database system. To
achieve persistence, all data is kept in memory, however, the
transactions which changes the database state, are send to
disk and stored in a transaction log. Transaction log can be
seen in the figure 1.
Fig -1: In-memory database with persistence
Even though the disk is used for storing transaction logs
there will be no major impact on the performance as the
transactions are appended at the end of the transaction log
file which works in an append-only way. The hard disks
work pretty faster when working in this fashion (append-
only). Hard disks can be written with as fast as 100 MB/s
when the data is appended at the end of the file. If used
sequentially, hard disks can work at a very high speed,
however, the disks will be slower when data is accessed
randomly.
In case of solid-state drives, data access speeds will be much
higher as there will be no moving parts like disks. In these
devices, sequential access can be between 200-300 MB/s.
Considering these, if a transaction size is of 100 bytes, then
using solid-state drives, it can be a million transactions per
second. This indicates, disks can never be a hindrance in for
in-memory database systems. In-memorydatabasesystems,
disk is not used when there no changes in the data and hence
in case of changes in data, disk is used in the fastest possible
way.
3. The Cloud and In-memory database systems
Cloud can be used to provide an excellent environment for
efficient use of in-memory database systems. Organizations
need not spend on purchasing large amount of primary
storage, instead they can use the services provided in the
cloud environment. In a cloud based environment failures
can be prevented due to the availability of the redundant
hosts and virtual machines. In such an environment, any
failure due to RAM will not lead to a data loss. To achieve
maximum performance gains, can combine in-memory
database systems with cloud based environment.
3.1 Oracle Database In-Memory
Performance enhancements of analytical queries are
achieved using the in-memorydatabaseaspectinOracle18c.
This feature is very useful in taking real-time business
decisions in an organization. Usingthisaspect,organizations
improve productivity and increase competitiveness. It can
useful in accelerating the data warehouse as well as OLTP
databases. Oracle 18c supports dual-format architecture
wherein data is maintained in a row-format for OLTP
operations and in a column-format for analytical processing.
The four key aspects of Oracle in-memory database which
enables orders of magnitude fasteranalyticqueryprocessing
are [2]:

• Compressed columnar storage
• Vector processing
• In-memory storage indexes
• In-memory optimized joins and reporting
Figure: Oracle dual-format architecture
3.2 In-Memory Column Store
One of the key aspects in the in-memory database is in-
memory column store. Classical relational databasesystems
store data in a row or column format. In these systems data
in both the memory and the disk are stored in the same
format. Data stored in a row format gives optimum
performance for transaction processing applications
wherein updating a some of the columnsina fewrowsneeds
to access and modify small number of blocks. Using the
same approach for analytical processing results in a poorer
performance as the analytical workloads need to access few
columns, however, scans the entire data set.
Using columnar format is efficient for analytical workloads.
In a columnar database, columns but not rows are stored
continuously. In a columnar format, as the columns are
stored separately, analytical query needs to access only the
required columns thereby avoiding the scan of data whichis
not required. For example, a reporttocomputeandshow the
sales total by city can quickly process large number of rows,
however, accesses only a few columns.
4. The Oracle Database In-Memory Solution
Most of the database vendors give one option to the
customers to either choose the columnar or the row format.
In such cases, if the data format is chosen as columnar, then
the columnar format is applicable in both the disk as well as
the memory. This indicates advantagesof boththecolumnar
and row formats cannot be obtained at the same time. In
case of databases which have both OLTP and data analytical
needs, performance problems will arise due to the single
format.
The Oracle in-memory aspect provides a complete solution
for both the analytical queries and mixed-use databases.
Oracle’s in-memoryaspectslikeIMcolumnstore,improvised
query optimization techniques, and availabilityprovidesthe
solution for faster executionofdata analytical queriesofdata
warehouses without sacrificing the performance
requirements of OLTP applications. The feature of
compressed columnar format enables faster scan, queries,
aggregates and joins. The IM column store is for improving
performance of analytical queries, the dual-memory format
architecture can indirectly improve OLTP performance. The
in-memory feature in Oracle database is easy to implement
and requires no changes in the application.
Scans done using the in-memory columnar format are faster
than the row-format due to the following reasons [1]:
 Avoids the overhead due to buffer cache.
 Only the required columns are scanned avoiding to
scan entire rows of data.
 In the in-memory column format, compression is
done to speed-up the scans.
 Vector scans by a CPU core are orders of magnitude
faster than row scans.
For example suppose a user executes the following ad hoc
query:
select inv_no, cust_name, supp_no
from sales_invoice
where item_no between 101 and 151;
In case of traditional database systems using the buffer
cache, data is fetched using index to find the item_no,
database uses the rowids to fetch and transfertherowsfrom
disk into the buffer cache and then discard the unneeded
data. Data scans in row formats require multiple CPU
instructions which may have an impact on the performance.
However, data scan using columnar format pipelines only
the required columns to the CPU, increasing the
performance.
5. CONCLUSION
In this research paper, traditional and in-memory database
systems are compared. To achieve performance gains of
data analytical queries and decision makingrequirementsof
the organization, Database Administrators(DBA)canusein-
memory database systems. In-memory database systems
also work efficiently for OLTP systems, DBAs must
implement relevant data format i.e. row or columnar as per
the requirements.
REFERENCES
[1] Ashdown, L. (2019) Oracle ® Database: Database In-
Memory Guide. Oracle Corporation
[2] Lahiri, T. (2015) ‘When to Use Oracle Database In-
Memory’. Oracle Corporation [online]
[3] Plattner, H. (2013) A Course In-Memory Data
Management. Springer
[4] D’Souza, S. (2012) In-memory database technology
gains ground, but challenges remain [online] available
from < http://www.computerweekly.com/feature/In-

memory-database-technology-gains-ground-but-
challenges-remain [2 June 2019]
[5] Mullins, C.S. (2015) How to determine if an in-memory
DBMS is right for your company [online] available
from <
https://searchdatamanagement.techtarget.com/featu
re/How-to-determine-if-an-in-memory-DBMS-is-
right-for-your-company> [ 25 June 2019]
[6] Mullins, C.S. (2017) What is an In-Memory Database
System? [online] available from <
http://www.dbta.com/ Columns/DBA-Corner/What-is-
an-In-Memory-Database-System-119241.aspx> [ 2 July
2019]
[7] Anikin, D (2017) Choosing BetweenanIn-Memoryanda
Traditional DBMS [online] available from <
https://dzone.com/ articles/when-and-why-i-use-an-in-
memory-database-or-a-trad> [13 July 2019]
[8] Anikin, D (2017) What an In-Memory Database is and
How it Persists Data Efficiently [online] availablefrom<
https://dzone.com/articles/what-an-in-memory-
database-is-and-how-it-persists> [25 July 2019]

IRJET- Improving Performance of Data Analytical Queries using In-Memory Database Systems

More Related Content

Similar to IRJET- Improving Performance of Data Analytical Queries using In-Memory Database Systems (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Improving Performance of Data Analytical Queries using In-Memory Database Systems