SSD vs HDD

  • perf-test.com need your contributions to build up a strong repository of performance engineering resources.

vineetmishra

Administrator
Aug 14, 2014
12
0
1
I have an application with very high batch database writes, my manager is thinking of using solid state drives to gain some performance and faster database writes. He has asked me to do a quick poc to see how much performance gain will we be getting.

Can someone give me a ballpark estimates on how much performance gain we can expect after moving the database on SSD drives. I am assuming a 5x - 7x gain
 
Vineet, SSD performance (when compared to normal HDD) depends upon many parameters like

1. No. of Outstanding I/O (or Queue depth)
2. Request Size
3. Read/Write ratio
4. Transfer ratio (Random/Sequential)

However even if we take a rough estimate, i will be not very optimistic on 5x improvements in disk IOPS. Since its a write operation i would assume that you may get a 2x improvement (Approximately) or even lower. I can share with you some benchmark numbers that we did on a similar application.
 
anmoldubey said:
Vineet, SSD performance (when compared to normal HDD) depends upon many parameters like

1. No. of Outstanding I/O (or Queue depth)
2. Request Size
3. Read/Write ratio
4. Transfer ratio (Random/Sequential)

However even if we take a rough estimate, i will be not very optimistic on 5x improvements in disk IOPS. Since its a write operation i would assume that you may get a 2x improvement (Approximately) or even lower. I can share with you some benchmark numbers that we did on a similar application.


Thanks Anmol. Any kind of data would definitely help us make a decision.
 
vineetmishra said:
anmoldubey said:
Vineet, SSD performance (when compared to normal HDD) depends upon many parameters like

1. No. of Outstanding I/O (or Queue depth)
2. Request Size
3. Read/Write ratio
4. Transfer ratio (Random/Sequential)

However even if we take a rough estimate, i will be not very optimistic on 5x improvements in disk IOPS. Since its a write operation i would assume that you may get a 2x improvement (Approximately) or even lower. I can share with you some benchmark numbers that we did on a similar application.


Thanks Anmol. Any kind of data would definitely help us make a decision.


IOMETER was used for doing this benchmark.The main reason for this being it is an open source tool and is readily available (You can download it from http://www.iometer.org ).

We observed following performance benifits for different workloads

Workload Performance factor $ per GB factor


OLTP - 15.84 X
Backup - 1.74 X
Restore - 2.45 X
File copy - 9.54 X
Database transaction - 15.23 X
 
Much hype is given to SSD performance and SSD is considered to be the next generation storage devices. The objective of our testing was to test the performance of an SSD in a real-world environment against workloads like OLTP, DSS etc., and to compare their performance with the traditional hard disk drives.

Tool used for bench marking - For evaluating the disk I/O performance, the most widely used tool is Iometer. The main reason for this being it is an open source tool and is readily available (You can download it from http://www.iometer.org ). The other reason is that, with Iometer almost any type workload mix can be generated and tested on the drives

Drive Specifications – SSD

Cost – 9000$, Make – HP, Interface – SAS,

Drive Specifications – HDD

Cost – 500$ , Make – HP, Interface – SAS

RESULTS

We simulated various commonly observed workloads like database, ftp, backup etc and compared SSD performance against HDD performance and we were surprised to see the difference in the results.

upload_2015-4-30_0-57-11.png
 
Sequential writes will be quite fast even in normal disks. Random reads and writes would be faster in SSDs. Hence OLTP and Database Transactions are much more faster in SSD than Backup, Restore and File Copy. For Random writes and reads use SSDs for sequential writes and using SSD will not give much benefit.