ABC Company

V7000 G1 Storage Assessment & Capacity Planning

By Nguyen Van Hai ([email protected] )

Client Technical Specialist, IBM Vietnam

04/2016
IBM Storage Systems

Agenda

Assessment & Capacity Planning Objectives

Sizing Methodology and Assumptions

Findings and Analyses

Capacity Planning for the next 3 years: 2016 – 2019

Assumption growth rate per year ~ 20%

Recommendations

2 © 2016 IBM Corporation

IBM Storage Systems

Assessment & Capacity Planning Objectives

The objectives of this Assessment & Capacity Planning for Storage System are

1. To analyze the current Storage System resources utilization, and perform

a capacity planning for the next 3 years (2016-2019)

2. To provide our findings and recommendations in order to

help improve the performance of current system set-up
recommend possible upgrade options

3. To open up a discussion between ABC company IT Team and IBM – other

requirements, pain points, etc., and how IBM can help.

3 © 2016 IBM Corporation

IBM Storage Systems

Sizing Methodology and Assumptions

1. Sizing is based current system resource utilizations, map against projected growth rate

2. Current system utilization data is derived storage performance data using Stats files
reports from 22 to 28, Mar 2016.
– Disk storage system: IBM Storwize V7000 G1 (Cluster_192.168.103.8)

3. Since each Storage Pool has one or more Managed Disks (Mdisk) and each Mdisk
has different peak overall IO rate (IOPS), with specific Read/Write ratio, cache hit,
block size… and each Mdisk also has one or more Volumes with different peak
overall response time (ms), so:
– We had take the peak over IO rate (IOPS) of each Storage Pool to do
performance analyze and capacity planning

4. Rule of storage system resources utilization is less than 70%; Overall response
times (front end - volumes) needs to be 5-15ms; Backend response times (Managed
Disks/mdisks) can usually operate at 20- 25ms, unless the hit ratio is poor, backend
write response time can be even higher, generally up to 80ms.

4 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage System

CPU utilization of two nodes (canister)

Analysis
The workload on the two storage nodes (canisters) were not balanced. Almost the workload is on Node2. Need
to check the preferred node (for your volumes control) and re-balance the workload to Node1 if any.

5 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall response time (ms) of the Volumes: T24_Arc, T24_DATA_Simu1, T24_DATA_
PreArc1, T24_DATA_PreArc4

Analysis
Almost overall response time of these volumes are less than 10ms.
So, there is no performance issue on these volume.

6 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall response time (ms) of the Volume T24_DATA

Analysis
• This volume is for T24 Jbase production volume and it is an important volume. We will analyze more detail on
this volume (& Pool_T24_DATA as well) during this report.
• Almost overall response time of T24_DATA are less than 10ms, it’s ok. Only at 19:17” 3/26/2016 it was
20.84ms which over the best practice for volume response time (<15ms). From our point of view, the
T24_DATA volume is ok for production system during business hours at this time. We will have more analyze
the Storage pool Pool_T24_DATA during the report.

7 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall response time (ms) of the Volumes: T24_DATA_Simu2, T24_DATA_Simu3,
T24_DATA_Simu4, T24_DATA_ PreArc2

Analysis
Almost overall response time of these volumes are bigger than 15ms.
Specially for T24_DATA_Simu3 was 959ms and T24_DATA_Simu4 was 111 ms at the peak time!!! It’s too big latency on
theses volume at the peak workload time.
• However, these volumes are for Simulation test and Pre-archiving only and they were not using open. We will have the full
picture on these on the end of the report.

8 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall response time (ms) of the Volumes: p7_TJLOG, FC1_DATA, p8_TJLOG, FC2_DATA

Analysis
Almost overall response time of these volumes are bigger than 15ms.
Specially for p8_TJLOG was 411ms and FC2_DATA was 171 ms!!! It’s too big latency on theses volume at the peak
workload time.
• However, these volumes are for Log and FlashCopy purposes only and belong to Pool_T24_TJ_FC storage Pool, not the
same with Production Data Storage Pool. Similar for the volume belong to Pool_T24_Backup_TSM and Pool_USER_DATA.
We will focus the analyze on Pool_T24_DATA for Jbase Production only.

9 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Mdisk0, Mdisk1, Mdisk2, Mdisk3 (& Mdisk4) (SAS 300GB 15k)

Analysis
• The peak I/O rate were < 700 IOPS for these Mdisks and almost the time the IOPS around 50 to 150 IOPS.
• One point you can see at the time 3/26/2016 11:22” all these volumes had the peak I/O workload. So, that
time should be a special time regarding to your business activity which has heavily workload need to be
attention.
• We will have more analyze about these Mdisks (Maximum IOPS capability) during our report.

10 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Mdisk5 (SSD drives) for Easy Tier function

Analysis
The peak I/O rate was 5827.27 IOPS and almost the time the IOPS around 1000 to 3500 IOPS.
This is a SSD Mdisk which was used for Easy Tier function on the Pool_T24_DATA Production storage Pool.
In this system, the Easy Tier function with SSD drives was worked very effective for hot data and improved the
overall system performance as well. We will have the Easy Tier function activity analyze in detail in next few slides.

11 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Mdisk13 (SSD drives) for Easy Tier function

Analysis
The peak I/O rate was 6100.31 IOPS and almost the time the IOPS around 1000 to 4000 IOPS.
Similar with Mdisk5, this is a SSD Mdisk also which was used for Easy Tier function on the Pool_T24_DATA Production
storage Pool. And the Easy Tier function with SSD drives was also worked very effective for hot data and improved
the overall system performance as well. We will have the Easy Tier function activity analyze in detail in next few slides.

12 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Mdisk7 (SAS 10k drives)

Analysis
The peak I/O rate was 1245.4 IOPS and almost the time the IOPS around 400 to 900 IOPS.
This storage pool is for Log and FlashCopy only and was running with SAS 10k drives but it was consumed the storage
system resources (controller cpu, cache, system bus…) in the overall system performance perspective. We will have a
total system resources utilization point of view when we combine with other storage pools in this system

13 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Mdisk6 (SAS 10k drives)

Analysis
The peak I/O rate was 1149.2 IOPS and almost the time the IOPS around 200 to 1000 IOPS.
This storage pool is for Backup only and was running with SAS 10k drives and it was consumed the storage system
resources (controller cpu, cache, system bus…) in the overall system performance perspective. We will have a total
system resources utilization point of view when we combine with other storage pools in this system

14 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Mdisk8 (SAS 10k drives)

Analysis
The peak I/O rate was 581.59 IOPS and almost the time the IOPS around 100 to 400 IOPS.
This storage pool is for User data only and was running with SAS 10k drives and it was consumed the storage system
resources (controller cpu, cache, system bus…) in the overall system performance perspective. We will have a total
system resources utilization point of view when we combine with other storage pools in this system

15 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Storage Pool Pool_T24_TJ_FC

Analysis
The I/O rate was 828.62 IOPS on Storage Pool Pool_T24_TJ_FC (including Mdisk7 only) at 20:07pm 3/22/2016.
We will combine this IOPS number with others storage pools at the same time to have the Total IOPS of the storage
system at the peak workload. This parameter will help us when we consider for the next 3 years planning as well.

16 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Storage Pool Pool_T24_BACKUP_TSM

Analysis
The I/O rate was 3.8 IOPS on Storage Pool Pool_T24_Backup_TSM (including Mdisk6 only) at 20:07pm 3/22/2016.
We will combine this IOPS number with others storage pools at the same time to have the Total IOPS of the storage
system at the peak workload. This parameter will help us when we consider for the next 3 years planning as well.

17 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Storage Pool Pool_USER_DATA

Analysis
The I/O rate was 0.27 IOPS on Storage Pool Pool_User_Data (including Mdisk8 only) at 20:07pm 3/22/2016.
We will combine this IOPS number with others storage pools at the same time to have the Total IOPS of the storage
system at the peak workload. This parameter will help us when we consider for the next 3 years planning as well.

18 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage Systems

Overall IO rate (IO/s) of the Storage Pool Pool_T24_DATA

Analysis
The peak I/O rate was 10652.92 IOPS on Storage Pool Pool_T24_DATA (including Mdisk0, Mdisk1, Mdisk2, Mdisk3,
Mdisk4, Mdisk5 and Mdisk13) at 20:07pm 3/22/2016. This is the peak workload time of this Storage Pool.
We will analyze more detail this information on the next few slides.

19 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage System

Analysis
 We saw that the total IOPS maximum (at 3/22/2016 20:07”) put on storage pool Pool_T24_DATA was 10652.9 IOPS.
 With the data collected at that time, we used Disk Magic Tool to calculate and see the system utilization will be as below:

You can see, with Easy Tier function was worked very effective so all system parameters utilization were
lower than 25%. It’s good. In there, the Highest SAS 15k Drive Utilization is 24.7%. Next slide, we will give
you more detail how about Easy Tier function was worked on this system.

20 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage System – SSD & Easy Tier

Analysis
We used Storage Tier Analysis Tool (STAT) to analyze the efficient of two SSD Mdisks (Mdisk5 and Mdisk13) and Easy Tier
function on this system and below is the result:

• We can see, we have total 10 volumes on Pool_T24_DATA but almost the Hot data was put on the T24_DATA volume (742GB of 750GB)
• The current storage system performance will be improving about 7.1% ~ 27.1% if we add 1x900GB SSD array RAID5 into this storage
pool as the recommendation from the STAT Tool.
• Moreover, we need to combine with 3 years planning requirements and growth rate per year for the capacity planning as well. See next
some slides on this.

21 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage System

Analysis
• We calculate the maximum IOPS of current storage system at the peak workload time (3/22/2016 22:07”) = total
IOPS of all Storage Pools at that time = 10652 + 828 + 3.8 + 0.2 = 11484 IOPS
• With assumption growth rate per year ~ 20%. We calculated the IOPS will be growth plan from now on for the next 3
years as below:

Below is the chart show that this storage system will be overloaded once the IOPS maximum reach to 19844.

• Based on the ABC’s 3 years planning (will be moving the production database from Jbase to Oracle with triple usable data)
and leveraging the Test/Dev/Reporting as well as the Backup/FlashCopy requirements.
• Combined with STAT Tool recommendation to add on 1x900GB SSD RAID-5, we recommend ABC to consider upgrade
your current storage system (V7K G1) to meet the growth plan for the next 3 years via add on more SAS drives (15k and 10k)
and SSD drives on this or you can consider to buy the IBM FlashSystem Storage to get highest performance for your
production system and utilize current storage system for Test/Dev/Reporting as well as backup system. See the next slide for
the proposed configuration and storage system resources utilization calculation.

22 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage System

Analysis
Proposed storage system configuration for the next 3 years planning

• In term of the performance utilization and improvement for the next 3 years, we proposed to add on some arrays as below:
 1x900GB SSD RAID-5 to the storage pool Pool_T24_DATA to improve Easy Tier function as STAT Tool recommendation
 1x4200GB SAS 15k 2.5” Mdisk (1xRAID-5) to the storage pool Pool_T24_DATA for Oracle database in Production
system
 2x4200GB SAS 10k 2.5” Mdisk (2xRAID-5, 8400GB in total) to the storage pool Pool_T24_TJ_FC for FlashCopy and Log
 For Backup and User data, you can add more SAS drives for these purpose depend on your requirements
• See next slide for the system storage performance & resources utilization for the proposed configuration.

23 © 2016 IBM Corporation

IBM Storage Systems

Findings and Analyses: Storage System

Analysis
Proposed storage system configuration performance & utilization estimation with the peak workload up to 19844 IOPS
and the usable capacity for Pool_T24_DATA will be ~ 12000GB (12TB)

• You can see the total system resources utilization should be lower than 70% in general and the IOPS maximum which can be
support by this system ~ 19844 IOPS.
• Moreover, ABC can consider to deploy your system on IBM FlashSystem storage which has very good performance and lowest
latency for OLTP/OLAP workload. With IBM V9000 FlashSystem storage you can get the highest performance from Flash
technology as well as re-utilize your current storage system (V7K G1) for Testing/Developing/Reporting and Backup System with
External virtualization feature from this product.

24 © 2016 IBM Corporation

IBM Storage Systems

Recommendation: Storage System

Based on the finding, analysis and capacity planning results on previous pages, we
recommended for the optimization and system upgrade as below:
Need to check the Preferred Node of each Volume and re-balance the
workload over two nodes (canister) for better CPU Utilization of the storage
system
Need to upgrade the current storage system firmware (7.2) to latest version
(7.5 +) to improve the performance and system stable
Should choose the best performance Mdisk (RAID group) with 8 hdds per
Mdisk in RAID-10 or 5. Consider to implement DRAID technology.
Need to consider to add more SAS drives and SSD drives to existing Storage
Pools to adapt with future growth plan
Recommended to add more drives to current storage system for the next 3
years plan with growth rate 20% per year can be:
 7 x 200GB SSD (for Easy Tier function improvement) and
 9 x 600GB SAS 15k (for capacity growth with Oracle database)
 17x600GB SAS 10k (for FlashCopy and Log purpose)
Recommend ABC frequently to use the Storage Tier Advisor Tool (STAT) to
check how much hot data they have and how many SSD drives need to be
add on to improve the storage system performance at all
Other option: ABC can consider to put their system on IBM FlashSystem and
re-utilize current storage system for Test/Dev/Reporting & Backup as well.

25 © 2016 IBM Corporation

IBM Storage Systems

Questions ?

26 © 2016 IBM Corporation