Quantitative Performance Evaluation of Cloud-Based MySQL (Relational) Vs. MongoDB (NoSQL) Database with YCSB

Data Storage and Management Project
on
Quantitative Performance Evaluation of Cloud-Based MySQL
(Relational) Vs. MongoDB (NoSQL) Database with YCSB
Darshankumar Vinubhai Gorasiya
x18134751
MSc/PGDip Data Analytics – 2019/20
Submitted to: Prof. Iqbal Muhammad

Quantitative Performance Evaluation of Cloud-Based
MySQL (Relational) Vs. MongoDB (NoSQL)
Database with YCSB
Darshankumar Vinubhai Gorasiya
x18134751
April 22, 2019
Abstract
NoSQL was built to satisfy the need of Big Data applications. Modern applica-
tion demands a database management system that can deal with large volumes of
rapidly changing data with geo-distributed architecture supporting easy scale-out.
However, there is a substantial number of business applications running on low-
volume environments having structured and gradually changing information and
data isn’t growing exponentially. NoSQL (schema-less) structure does not hold the
ACID properties (Atomicity, Consistency, Isolation, Durability) and loses the ref-
erential trustworthiness and consistency assurance of data which conventional SQL
(relational) databases provide and it is vital to numerous mission-critical small-scale
applications.
Migration of existing highly structured database to NoSQL for better perfor-
mance has been trending among medium scale businesses however it is an expansive
and complex process. This report replicates low-volume and high-volume applica-
tion operations on NoSQL and Traditional SQL database with Yahoo! Cloud Serv-
ing Benchmark (YCSB) for analyzing quantitative performance differences between
NoSQL document-oriented database MongoDB and relational database MySQL.
1 Introduction
The traditional databases like MySQL, Microsoft SQL Server, Oracle Database, Post-
greSQL or IBM DB2 are still being utilised by large number of organisations and in-
dividuals for the purpose of holding relational data due to its ease of use and need for
routine analysis, industry-wide support, SQL standards, reliability of transactions and
large developer community support. The Digital revolution and increases in embedded
devices generating data and growing applications of ML (AI) have caused a constant rise
of distributed NoSQL (Not Only SQL) databases implementation demand which com-
promises consistency while provides availability and flexibility of scaling horizontally as
mention in Huang Y. (2014).
The Stack Overflow Developer Survey StackExchange (2018), the worlds largest de-
veloper survey with over 100,000 respondents indicates that MySQL is the most famous
database and still widely being used with the greater part of the respondents (58.7%) con-
firms utilizing MySQL where the same survey unveils that MongoDB is the most wanted
1

database. There is growing demand for NoSQL like better performance and SQL like
flexibility, consistency and strong transactional support in medium and high volume tra-
ditional databases application and large NoSQL commercial and open source databases
vendors are increasingly working on with different design approach to address the urging.
MongoDB is one of the vendor to provide support for multi-document ACID transactions
in version 4.0 of its NoSQL database.
In the Big Data world major trade-off between all the different distributed database
can be categorised based on CAP Theorem and its attributes are as follow,
1. Consistency -
Data is consistent throughout when a transaction begins and when it closes and all
the user should be able to see same information.
2. Availability -
At least one copy of data should be always available to users whenever requested.
3. Partition Tolerance -
Irrespective of deployed platform system performance should be tolerance to net-
work partitions.
Any distributed database management system cannot achieve more than 2 attributes
of CAP at a time. MySQL prioritises Consistency & Availability (CA) where loses on
partition tolerance and MongoDB relays on consistency & partition (CP).
In this study we will compare performance of MySQL (Consistency & Availability -
CA) with MongoDB (consistency & partition - CP). Yahoo! Cloud Serving Benchmark
(YCSB) automated workloads used for quantitative comparison with large and small data
volume.
2 Key Characteristics of The Data Storage Manage-
ment Systems
2.1 NoSQL: MongoDB
1. Aggregation Framework (MapReduce) -
Aggregation operations in MongoDB process data records and returns computed
results. It is built on the idea of information handling pipelines where documents are
given as input and gets aggregated data as result. MongoDB also provides efficient
map-reduce operation support for aggregation for efficient data summarization.
2. Horizontal Scalability (Sharding) -
MongoDB bolsters horizontal scaling through Sharding, disseminating data over
multiple machines which in turn helps to facilitate high throughput on high volume
datasets. Sharding permits adding additional instances to expand the capacity of
the database as and when required.
3. Capped Collections -
MongoDB supports the creation of broadly utilized capped collections where we can
confine the collection’s maximum size. This collection follows the kind of circular
queue concept where once collection size reaches max dispensed size limit, it starts
making space for new data by supplanting the oldest document in the collection.
2

4. Support for Multiple Storage Engines -
The storage engine in MongoDB guarantees how data is being put away in-memory
and on disk. MongoDB provides freedom of picking different storage engines as per
the application requirement as any single engine might not be the most efficient
solution for all sort of uses.
5. Documents Indexing -
Indexes stores the collections datasets in a structure that is easy to traverse. Mon-
goDB supports single, compound, multikey, geospatial and hashed indexes on data.
It helps discover documents matching query requirements without executing the
entire collection scan.
6. Master Slave Replication (Replica Sets) -
MongoDB underpins replication by making various copies of data over different
locations to protect data when the database suffers any loss. Concept replica set
in MongoDB gives consistency for all read operations from the primary node as at
least one redundant copy will be accessible in secondary node copies.
7. Load Balancing (Read-Write) -
MongoDB scalable architecture is upheld by the load balancer to dynamically bal-
ance the operation load of queries and manages balanced documents spread over
multiple nodes for reads and write tasks.
8. High Performance -
MongoDB uses the binary object BSON for storing data in documents. Binary
improves the overall performance of MongoDB with support of replica sets, Sharding
and load balancer.
9. Schema-less Database -
MongoDB is Schema-less database where different set of data and fields with dif-
ferent types can be dynamically stored and it makes MongoDB schema migrations
simple in large organizations. Any sorts of data can be adjusted in the database at
ease.
2.2 Relational: MySQL
1. Robust Transactional Support (ACID) -
MySQL holds amongst the most robust transactional database engines. It bolsters
row-level locking, multi-version transaction and server-enforced referential trust-
worthiness to achieve full ACID (atomic, consistent, isolated, durable transaction)
qualities.
2. Replication (Built-in) -
MySQL is one of few databases who guarantees nonstop uptime as a profoundly
available solution. It can be implemented by configuring cluster specialized servers
and master-slave replication architecture. There are also additionally third-party
vendors in the market giving out of box availability solution for MySQL.
3. Highly Secure System -
MySQL offers an exceedingly secure security framework that reinforces both clients
3

based and host-based verification. It likewise offers support of SSH/SSL for secure
connections to the database with object-level privilege framework to accomplish
high data security.
4. SQL Compatibility -
MySQL is one of the most established and popular open source relational database
systems. It supports widely accepted ANSI standard SQL where large developer
community utilizing it. MySQL additionally gives a range of MySQL connectors and
drivers like ODBC and JDBC to support seamless external application integration
and connectivity.
5. Fully relational DBMS -
MySQL is a Relational database management system and it isolates or groups
related data with help of logical database structure like a table, columns and views
and relates them by linking tables with each other in the database while physical
data is composed into a disk as files.
6. Platform Independence -
MySQL is platform independence and can be used on the number of operating
systems without worrying about compatibility issues caused due to the difference
in operating system.
7. Multi-Threaded SQL Server -
MySQL server supports multi threaded task management which allows users pro-
cess to be distributed with multiple small threads on different processing units of
machine.
8. Supports Geographic Information System (GIS) functions -
MySQL provides functions to process GIS (geographic information system) data
objects. Objects having spatial attributes like geographical coordinates. Any data
having more than one dimension can be treated as a spatial entity
3 Database Architectures
3.1 MongoDB
MongoDB is constructed based on the design philosophy to utilize capabilities of rela-
tional databases with the addition of best of innovative NoSQL technologies. Its unique
performance capabilities make to the choice of many organizations as a storage solution
for Big Data applications and advanced real-time analytics MongoDB (2019).
1. Storage Engines -
MongoDB storage engine enables utilization of the best of various storage engines.
There are four predominantly supported storage engines accessible and all four can
work in a single replica set MongoDB. It comes with default WiredTiger storage
engine that furnishes all-around performance with local compression support. En-
crypted storage engine ensures sensitive data protection without influencing the
performance and the in-memory storage engine (ISE) provides wide support for
real-time analytics applications.
4

Figure 1: MongoDB Architecture with Sharding
2. Sharding -
MongoDB supports horizontal scale-out architecture to address hardware limitation
for databases by utilizing commodity hardware to distribute data among those
hardware and this technique is called Sharding and those commodity hardware
nodes are called shards. In case of addition or removal of shards or exceptional
data growth, Sharding is automatically overseen by MongoDB.
3. Replica Sets -
As part of Sharding implementation, MongoDB makes numerous copies of data on
different hardware, those copies are called Replica Sets. It is accomplished with
native replication support where one replica set performs the role of a primary set
and other act as a secondary set. All the read and write operations to a replica set
are made through the primary set and then copies are distributed on secondary sets.
In case of primary sets downtime, any of the secondary sets gets elected as primary
to take request from the application. Sets are self-healing shards which means that
it automatically takes care of failure and reduces the database downtime.
4. Query Model -
MongoDB query model consists query router, it provides transparency which implies
that regardless of the number of shards the query code has to be triggered the same
way. It optimizes query performance with help of caching, query plan and indexes.
It also provides support for query optimization at run time by utilizing more than
one index.
3.2 MySQL
MySQL standout amongst the most adaptable and flexible relational database architec-
ture and its behaviour can be tuned dependent on optimum utilization of the diverse
components. It is mainly divided into 3 layers, first is where all the connection manage-
ment happens, second is core processing unite of architecture where all the processing
5

takes place and third is data engines where data storage and retrieval takes place Oracle
(2019).
Figure 2: MySQL Architecture with core components
1. Client -
The client is where all the application accessing MySQL lives and access server for
storage and retrieval of information. It very well may be accomplished by using
utility driver provided to connect server thread managers like ODBC or JDBC.
2. Connection Management -
Every client connection gets one dedicated thread in the server and within that
connection, the query gets executed for the client. The server also manages caches
threads for reducing overhead for connection management. The server allows access
to each client connection request with username, password and host-based authen-
tication.
3. Optimization/Execution -
MySQL as part of query execution takes request from thread received from the
client and parses queries to create a parse tree which is similar to execution plan and
then optimizes it by rewriting query busing variety of methods utilising indexes and
cache. The storage engine is aﬀected by how the server optimizes the user request
for best performance, hence optimizer in server consults with storage engines to
generate the most agent execution plan.
4. Metadata Cache -
Cache in MySQL assumes the indispensable role for performance and it is mainly
6

used in the form of Query and Key cache. When a client sends similar query then
the server does not parse and execute that identical query again it just extracts
execution plan from query cache whereas Key cache is used for storing table indexes
which later used by storage engines.
5. Storage Engine -
The storage engine is a crucial part of MySQL server as it is responsible for the
data management in the physical disks. It executes write and read operation on
the physical data files in disks. InnoDB is one the most broadly utilized and oracle
recommended default storage engine for almost all MySQL requirements.
4 Capabilities of the Database Management Systems
This section describes and compares MongoDB and MySQL on the scale of Scalability,
Availability, Reliability and Transaction Management as briefly discussed by Shah (2017).
4.1 Scalability, Availability and Reliability
As per the CAP theorem any distributed data management system cannot achieve more
than two attributes from CAP at the same time and same is the case for both of the
database systems used in this study. MongoDB priorities CP (consistency & partition)
which means that MongoDB design compromises on availability. While making sure
that data is consistently available for user requests by deprioritizing availability also with
flexibility on schema and inbuilt replica sets scalability can be easily achieved. Opposite
to it in MySQL technical architecture strictly follows CA (consistency and availability)
which means that data will be available on and consistence across node but on partition
tolerance hence may not better perform in case of partition between multiple nodes.
4.2 Transaction Management
MySQL is open source relational database management system which strictly follows
ACID property of transaction (Atomic, Consistent, Isolated and Durable) model. Which
is highly suitable for application highly relays on the transitions completeness. Which
is not readily available in MongoDB NoSQL architecture however it can be achieved
with newer version recognizing need of ACID in NoSQL framework but it is still requires
difficult implementations and complex configuration. MongoDB design follows BASE
(Basically Available, Soft-State and Eventual Consistency) characteristics there is lack
of atomic transactions but it focuses on availability most of the time by making multiple
copies over various physical disks and achieves eventual consistency by making sure data
is consistent across node at.
5 Findings from Literature Survey
The dramatic increase in Big Data applications has exploded the demand for better
performing, salable and flexible database management system. This demands in the
market have led to the explosion of vendors coming out with various innovative database
architecture and solutions and NoSQL is one of that concept that got the breakthrough
7

for the Big Data applications Panda et al. (2015). Many studies has happened which
rules out the idea that NoSQL can replace the Relational database due to its ACID
compliance and complex operations support. As rightly said in the study by Wei Sun
(October 28, 2013), that when contrasted with RDBMS, NoSQL frameworks are built
to address the interest of system that can process a large amount of data with high
adaptability, accessibility and faster performance while waving off the ACID compliance.
However, RDBMS solutions have been in place for many years and still being used
by many organizations. It becomes crucial to understand how both compare in perfor-
mance with the help of benchmarking data storage tools. There are many benchmarking
tools available in market. The study by Bermbach D. (2017) says that the most famous
benchmark standard used to be TPC however that was only used for the RDBMS. With
the advancement of Big Data and NoSQL applications that might not be the right choice.
At the moment the widely accepted choice for NoSQL and Relational database compar-
ison is Yahoo! Cloud Serving Benchmark (YCSB) as it does not require the system to
have a relational schema or transactional support hence easy integration of NoSQL. In
this project, we are using YCSB to evaluate the performance of Relational and NoSQL
database system on the cloud.
Lately, in NoSQL era YCSB has been used in multiple academic and institutional
studies, One such is an experiment by JeonDong-cheol et al. (2018) for the relative
examination of NoSQL with RDBMS dependent on medicinal data shows that for med-
ical information processing systems, NoSQL has better execution performance for large
volume data sets as compared to relational. However, there has been similar research
done by Abramova et al. (2015) for SQL or relational (MySQL) and NoSQL (MongoDB)
platforms for the purpose of performance evaluation on extracting decision support by
complex queries on both system and results indicate that NoSQL missed the mark amid
execution as MongoDB like NoSQL system built for simple key-value pair data retrieval
and insertion.
MySQL is one of the widely used Relational databases till date and MongoDB repre-
sents the best of NoSQL, Research by Aboutorabi et al. (2015) studies the performance
of the MySQL and MongoDB database to replicate the large system having e-commerce
data and it directs that the MongoDB performs way more efficiently as compared to
MySQL over multiple reads and write operation tests. However, there are also study
like R. Panda (2015) which opens discussion with the argument that performance bench-
marking also depends on different database implementation design and hence may require
various benchmarks before deriving the conclusion. Research by Cornelia Gyrdi (2016)
studying MySQL and MongoDB as a backend system that processes a large volume of
the user simultaneously accessing the system. It supports the previous argument that
MongoDB offers the best possible solution as far as the speed at which diverse tasks can
be performed in parallel is a concern.
This research papers and studies happened previously helps us get an overview of
what are all the possible options for the purpose of cloud-based databases benchmarking
and what were the results generated from YCSB for applications specific tests replicating
medical, e-commerce and large back end online systems. In this project, we are studying
the performance of MYSQL and MongoDB by reproducing application environment using
large (150000), small (50000) and in between size of data volume with YCSB update heavy
and read-modify heavy workloads.
8

6 Performance Test Plan
To evaluate test performance of two database systems we are using Yahoo! Cloud Serv-
ing Benchmark where one database is Relational (MYSQL) and second is NoSQL (Mon-
goDB). This test will extract results from low volume data sets to as high as 1,50,000
record data sets load to broaden coverage of applications.
1. Local Machine
• Processor: Intel(R) Cire(TM) i7 CPU @2.80GHz
• RAM: 16GB
• OS: Windows 10
• OS Type: x64bit based OS
• CPU Cores: Octa-Core (8)
2. Cloud Hosted Virtual Machine
• OS: Ubuntu 18 (Bionic)
• Cloud Host: Openstack
• Instance Name: x18134751-DSMPROJB
• Specs: m1.medium (2 VCPU, 40GB Disk)
• RAM: 4GB
• IP Addresses: 192.168.101.155 (Floating: 87.44.4.63)
3. MongoDB Conﬁguration
• Version: v3.2.10
• Database: ycsb
• Test Collection: usertable
4. MySQL Conﬁguration
• Version: MySQL Ver 14.14 Distrib 5.7.25
• Database: BenchTest
• Test Table: usertable
5. Benchmarking and Testharness
• Tool: Yahoo! Cloud Serving Benchmark
• Version: ycsb-0.15.0
• Testharness Setup: NCI provided Unix shell based bash script setup to auto-
mate YCSB test workloads
6. Workloads
• Workload A: To replicate environment of application having read and write
ratio of 50-50%.
9

• Workload F: To replicate environment of application having 50% read and 50%
Read-modify-write activities. Example of application, Where users entries or
records are read and modiﬁed by the users.
• Load and Run Operations:
1) 50,000
2) 1,00,000
3) 1,20,000
4) 1,30,000
5) 1,50,000
7. Visualization Tool
• Tool: Tableau Software
• Version: 2019.1.0
7 Evaluation and Results
To evaluate performance of the NoSQL and Relational database we have used Yahoo!
Cloud Serving Benchmark which allows to trigger load and run test with various ratio of
read and write operations to capture latency and throughput for the database system.
For this study we are going to evaluate performance with below two workloads of
YCSB 15,
• Workload A: Update heavy workload (Read/Update ratio: 50/50)
• Workload F: Read-modify-write workload (Read/Read-Modify-Write ratio: 50/50)
7.1 Workload A: Update Heavy Workload
With help of Testharness scripts, Workload A has been executed against ﬁve operation
sets (50000,100000,120000,130000,150000) to extract Read, Write operation performance.
7.1.1 Overall run operations throughput (ops/sec) comparison with opera-
tion counts
This section describes overall throughput (ops/sec) of both the database to analyze how
both performed when executed for multiple data sets.
Workload Database Total Operations Throughput(ops/sec) Total
Workload A MySQL 50000 13.5465377758752
Workload A MongoDB 50000 4762.35831983998
Workload A MySQL 100000 17.4918903223493
Workload A MySQL 120000 8.08997017632494
Workload A MySQL 130000 11.5190999081661
Workload A MySQL 150000 14.0697635154148
10

Figure 3: Overall Throughput (ops/sec) Vs Operation Counts
Overall it includes total run throughput of workload A (Insert,Read,Update). It is
visible in Figure 3 that MongoDB’s overall throughput is between rang from 3622 to 4808
operations per second where as MySQL ranges from 8 to 17 operations per second which is
way more less than of MongoDB. MongoDB gives best throughput when executed largest
data set (1,50,000) where as MySQL shows it’s best throughput at 1,00,000 operation
count.
7.1.2 Average latency (µs) comparison with total READ operations
This section describes Average latency (µs) for various data operations of READ activity.
Workload Database READ Operations Throughput(ops/sec) Total
Workload A MySQL 24917 299.836256371152
Workload A MySQL 49980 4132.02160864345
Workload A MySQL 59842 15605.8340630326
Workload A MySQL 65278 8123.6801678973
Workload A MySQL 75009 6417.72324654374
The bar graph in Figure 4 shows that for the Read operation of Workload A, MySQL
gives best performance for small data set and average latency keep on increasing with
11

Figure 4: Average latency (µs) Vs Total READ Operations
increase in operation count. It is important because this read operation performance
of MySQL is comparably close to NoSQL MongoDB’s read operation latency. MySQL
latency is highest for operation count 59842 whereas MongoDB stays stable across various
operation counts.
7.1.3 Average latency (µs) comparison with total INSERT operations
This comparison describes Average latency (µs) for Insert operation on various records
count.
Workload Database INSERT Operations Throughput(ops/sec) Total
Workload A MySQL 50000 84255.18922
Workload A MySQL 100000 143253.17659
Workload A MySQL 120000 229566.452766666
Workload A MySQL 130000 173113.661330769
Workload A MySQL 150000 136212.42236
12

Figure 5: Average latency (µs) Vs Total INSERT Operations
For the workload A Insert operation Figure 5 shows, Average latency performance of
MongoDB is comparatively having high latency at 100000 operation count and depicts
least latency for operation count of 1,50,000. MySQL’s latency is extremely high s com-
pared to MongoDB for workload A insert operations where worst latency is coming for
1,20,000 operations at 229,566 (µs).
7.1.4 Average latency (µs) comparison with total UPDATE operations
This section compares Average latency (µs) on various data sets for Update operations.
Workload Database UPDATE Operations Throughput(ops/sec) Total
Workload A MySQL 25083 146816.669497269
Workload A MySQL 50020 110136.949240303
Workload A MySQL 60158 231020.341500714
Workload A MySQL 64722 166152.64990266
Workload A MySQL 74991 135723.834953527
13

Figure 6: Average latency (µs) Vs Total UPDATE Operations
As depicted in Figure 6, Comparatively MongoDB stays way ahead of MySQL for all
the records count of update operation. however MongoDB’s average latency stays close
to constant but MySQl’s latency for update decrease after 60158 record operation where
MySQL performs worst for 60158 record update operation.
7.2 Workload F: Read, Modify & Write Workload
With Testharness automated job, Workload F has been executed against ﬁve operation
sets same as Workload A to extract Read, Write operation performance. In workload F
Read operation and Read-Modify-Write operations are divided in 1:1 ratio.
7.2.1 Overall run operations throughput (ops/sec) comparison with opera-
tion counts
In this section we will compare performance by overall throughput (ops/sec) of both
the database with number of operation executed on database. Operation includes Read,
Insert and Read-Modify-Write combination.
14

Workload Database Total Operations Throughput(ops/sec) Total
Workload F MySQL 50000 24.3703909498116
Workload F MongoDB 50000 3640.86506954052
Workload F MySQL 100000 9.86670281826579
Workload F MySQL 120000 11.993439588545
Workload F MySQL 130000 11.3786892555238
Workload F MySQL 150000 7.92883500570083
Figure 7: Overall Throughput (ops/sec) Vs Operation Counts
Figure 7 shows that MongoDB and MySQl’s overall throughput for all the run oper-
ations and it is visible from line chart that MySQL performs maximum 24 operation for
50,000 operation count however if compered with MongoDB’s throughput of 3641 for the
same number of operation, MongoDB stays way ahead in comparison. There is sudden
fall in performance visible for MySQL when increased number of operation from 50k to
100k.
7.2.2 Average latency (µs) comparison with total READ operations
This section compares Average latency (µs) on various data sets for READ operations of
Workload F.
15

Workload Database READ Operations Throughput(ops/sec) Total
Workload F MySQL 50000 282.90992
Workload F MySQL 120000 7655.24389166666
Workload F MySQL 130000 9051.26092307692
Workload F MySQL 150000 13623.7114666666
Figure 8: Average latency (µs) Vs Total READ Operations
For the Read operation of Workload F as in Figure 8 confirms findings in Workload A
that MySQl’s performance for Read operations is comparatively close to average latency
of MongoDB when operation count is 50k. Read latency in MySQL after increase in record
count to 100k suddenly increase close to 10k (µs) whereas MongoDB stays constant across
various operation counts.
7.2.3 Average latency (µs) comparison with total INSERT operations
It describes how MongoDB and MySQL Insert operation performance differs for YCSB
workload F on various data sets.
16

Workload Database INSERT Operations Throughput(ops/sec) Total
Workload F MySQL 100000 174048.88175
Workload F MySQL 120000 169414.859641666
Workload F MySQL 130000 172377.422107692
Workload F MySQL 150000 225773.478133333
Figure 9: Average latency (s) Vs Total INSERT Operations
Overall average latency for insert operation of Workload F in MySQL stays constantly
high where in MongoDB it is close to constant except for operation count 100K where
average latency is high as compare to other load of MongoDB at 252 (µs) as shown
in Figure 9. For workload F as well insert latency of MongoDB is trading high on
performance as compared to MySQL.
7.2.4 Average latency (µs) comparison with total READ-MODIFY-WRITE
operations
Read-Modify-Write operation hold weight of 50% for entire workload F and this section
compares performance of those operations.
17

Workload Database READ-MODIFY-WRITE Operations Throughput(ops/sec) Total
Workload F MySQL 24912 82038.9455282594
Workload F MySQL 49890 194606.567187813
Workload F MySQL 59928 158978.945885062
Workload F MySQL 64861 167095.747814557
Workload F MySQL 74861 238461.344251345
Figure 10: Average latency (µs) Vs Total READ-MODIFY-WRITE Operations
In workload F read-modify-write run, Average latency of MySQL for all the operation
counts stands extremely high between 80k to 300k whereas MongoDB’s latency is between
200 to 500. MongoDB’s latency stays almost constant irrespective of operation count
while MySQL indicates least latency when executed load for least operations and it
shows worst performance when triggered for largest test volume data set.
8 Conclusion and Discussion
To conclude the project, Comparing NoSQL and Relational database system on cloud,
we carried out various performance test on database with help of Yahoo! Cloud Serving
18

Benchmark where we used Workload A which replicates update heavy application work-
load with read and update ratio of 50/50 and Workload F that depicts performance of
read-modify-write application workload with read/read-modify-write ratio of 50/50. We
executed both workload on different five data volumes to test the performance with large
and small volume data set. As previously concluded in many studies MongoDB surpasses
way ahead in the test results as compared to MySQL on all the size of workloads except
for the read operation where MySQL and MongoDB performance is close for low volume
test data set. All the database on cloud were running on single node architecture for the
purpose of this study in the future work, we aim to study the performance fluctuation
of MySQL on shared multi-node cloud environment with MongoDB to further move the
study towards the conclusion.
References
Aboutorabi, S., Rezapour, M., Moradi, M. & Ghadiri, N. (2015), ‘Performance evaluation
of sql and mongodb databases for big e-commerce data’.
URL: https://doi.org/10.1109/CSICSSE.2015.7369245
Abramova, V., Bernardino, J. & Furtado, P. (2015), ‘Sql or nosql? performance and
scalability evaluation’, International Journal of Business Process Integration and Man-
agement 7, 314.
URL: https://doi.org/10.1504/IJBPIM.2015.073655
Bermbach D., Wittern E., T. S. (2017), ‘Getting started in cloud service benchmarking’,
Cloud Service Benchmarking 7.
URL: https://doi.org/10.1007/978-3-319-55483-914
Cornelia Gyrdi, Ioana Andrada Olah, R. G. L. B. (2016), ‘A comparative study between
the capabilities of mysql vs. mongodb as a back-end for an online platform’, Interna-
tional Journal of Advanced Computer Science and Applications 7(11).
URL: https://doi.org/10.14569/IJACSA.2016.071111
Huang Y., L. T. (2014), ‘Nosql database: A scalable, availability, high performance
storage for big data’, Pervasive Computing and the Networked World .
URL: https://doi.org/10.1007/978-3-319-09265-219
JeonDong-cheol, Mun, L. & HwangHeejoung (2018), ‘Performance analysis of rdbms and
mongodb through ycsb in medical data processing system based hl7 fhir’, Journal of
Korea Multimedia Society 21(8), 934–941.
URL: https://doi.org/10.9717/KMMS.2018.21.8.934
MongoDB, I. (2019), ‘Introduction to mongodb’.
URL: https://docs.mongodb.com/manual/introduction
Oracle (2019), ‘Mysql documentation’.
URL: https://dev.mysql.com/doc/
Panda, R., Erb, C., LeBeane, M., Ryoo, J. H. & John, L. K. (2015), ‘Performance
characterization of modern databases on out-of-order cpus’, pp. 114–121.
URL: https://doi.org/10.1109/SBAC-PAD.2015.31
19

R. Panda, C. Erb, M. L. J. H. R. L. K. J. (2015), ‘Performance characterization of modern
databases on out-of-order cpus’, pp. 114–121.
URL: https://doi.org/10.1109/SBAC-PAD.2015.31
Shah, M. (2017), ‘Mongodb vs mysql: A comparative study on databases’.
URL: https://www.simform.com/mongodb-vs-mysql-databases/
StackExchange (2018), ‘Developer survey results 2018’.
URL: https://insights.stackoverﬂow.com/survey/2018
Wei Sun, A. P. (October 28, 2013), ‘Benchmarking availability of large scale data storage
applications’, CS848 1.
20

Quantitative Performance Evaluation of Cloud-Based MySQL (Relational) Vs. MongoDB (NoSQL) Database with YCSB

More Related Content

What's hot

Similar to Quantitative Performance Evaluation of Cloud-Based MySQL (Relational) Vs. MongoDB (NoSQL) Database with YCSB

More from Darshan Gorasiya

Recently uploaded

Quantitative Performance Evaluation of Cloud-Based MySQL (Relational) Vs. MongoDB (NoSQL) Database with YCSB