This post is more than 5 years old
4 Operator
•
1.4K Posts
0
3020
May 22nd, 2012 07:00
How is the rule of thumb of disk performance reference value calculated?
Hi experts,
I have three questions regarding performance:
(1) I learned that for a 15k rpm FC drive, the rules of thumb for drive performance is IOPS = 180, bandwidth = 12MB/s. As i know, different I/O profile would make the drive deliver different result. Say if the workload is perfectly large sequential, the bandwidth may be doubled, even more, right?
(2) how is the above valued calculated and under what kind of I/O profile?
(3) For I/O profile that is mostly [large, sequential], then the response time should be increased quickly. Let's say for CIFS copying, does it care about response time? Who determines the response time requirements? CIFS protocol itself or explorer.exe for Windows when doing file copy for example?
thank you!!
rupal1
433 Posts
0
May 22nd, 2012 08:00
I tried giving you a answer for this by typing but trust me its not a 1, 2,3 line answer so would ask you to refer the doc
EMC Clariion Storage system Fundamentals and Availability which can be found in powerlink @ Home > Products > Hardware/Platforms > CLARiiON CX4 Series > White Papers
In this pdf start reading Chapter 8 from page 78.
You will understand this in minutes.
MIght help
Thanks
zhouzengchao
4 Operator
•
1.4K Posts
1
May 23rd, 2012 21:00
I think i found the answer:
For 15k rpm disk, the IOPS = (1/service time) * 1024ms/sec = 1/5.56 * 1024 = 184 IO/s
Bandwidth = IO size * IOPS, say we have Bandwidth = 12MB/s, then the I/O size = 12/184 * 1024 = 66.782, which means the value is caltulated under large I/O. That's is make sense as we always describe bandwidth by large sequential I/O.
Many thanks to rupal!!
rupal1
433 Posts
0
May 24th, 2012 01:00
Happy to help
SteveZhou
2 Intern
•
136 Posts
0
May 24th, 2012 02:00
One more question, the fomular is: IOPS = (1/service time) * 1024ms/sec
why 1024 is used? it doesn't make sense for me, i think it should be 1000, how do you think?
Madhug1
2 Intern
•
127 Posts
1
May 24th, 2012 03:00
Hi,
In EMC Clariion Storage system Fundamentals and Availability document, you may refer page 79 "A Common Calculation Error" as this would clarify your doubt.
Thanks
Madhu
SteveZhou
2 Intern
•
136 Posts
1
May 28th, 2012 20:00
thank you, Ganapa, but I don't think it was due to the difference with "network and bus speed".
According to the formular, we may got the following:
1 IO IOPS
--------------------------- = ----------------------------, IOPS = 1000(ms) / service time (ms) IO
Service Time (ms) 1000 (ms)
As we can see, the 1000 should be a timing unit, not a storage unit. Storage Unit is not make sense in this equation.
Please correct me, thank you!
zhouzengchao
4 Operator
•
1.4K Posts
0
May 29th, 2012 19:00
Same questions....
kelleg
4.5K Posts
2
May 30th, 2012 12:00
The number should be 1000 not 1024 - that was a typo in the Fundements documents. We'll try to get that fixed in the next release.
glen