Professional Documents
Culture Documents
Disclaimers
IBM has not formally reviewed this document. While effort has been made to verify the information, this document may contain errors. IBM makes no warranties or representations with respect to the content hereof and specifically disclaim any implied warranties of merchantability or fitness for any particular purpose. IBM assumes no responsibility for any errors that may appear in this document. The information contained in this document is subject to change without any notice. IBM reserves the right to make any such changes without obligation to notify any person of such revision or changes. IBM makes no commitment to keep the information contained herein up to date. Note: This presentation is intended for IBMers and IBM BPs only. As IBM has not formally reviewed this document it may be presented but should not be handed out to clients (especially not as ppt version). A pdf version of the presentation without slides 33, 52 to 54 and 79 may be suitable as hand-out for clients at one's own responsibility using one's own best judgement. If you have any suggestions or corrections please send comments to: gerosch@de.ibm.com
2 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: 2.5" & Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: 2.5" & Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
DS3000 Entry-level
DS5000 Midrange
Storwize V7000
XIV Enterprise
DS8000
Selecting a storage subsystem: entry-level, midrange or enterprise class support for host systems and interfaces overall capacity & growth considerations overall box performance advanced features and copy services price, costs / TCO, footprint, etc. needs to meet client & application requirements
5 2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Note: Results as of 6-26-2006. Source of information from Engenio and not confirmed by IBM. Performance results achieved under ideal circumstances in a benchmark test environment. Actual customer results will vary based on configuration and infrastructure components. The number of drives used for MB/s performance does not reflect an optimized test config. The number of drives required could be lower/higher.
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Note: Results as of 6-26-2006. Source of information from Engenio and not confirmed by IBM. Performance results achieved under ideal circumstances in a benchmark test environment. Actual customer results will vary based on configuration and infrastructure components. Drives were short-stroked to optimize for IOPs performance. Real-life may take more drives to achieve the numbers listed..
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
http://www.storageperformance.org
8 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: 2.5" & Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Application I/O performance: Efficient memory usage is key! Access to memory is >10000 times faster than disk access!
M E M O R y
S E R V E R
SAN
SCSI
Hardware
C A C H E
S T O R A G E
Hardware Setup
Storage subsystem Cache hit: < 1 ms Physical HDD: ~ 5...15 ms Storage I/O performance: Proper data placement is key!
2011 IBM Corporation
System Software
10 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
CPU
MEMORY
100ns = 0.000000100s
12
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
CPU
1 cycle := 1 second
MEMORY
1:40 minutes
SLOW
DISK
116 days
13
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
I/O
I/O
Transaction Processing A single end-user is capable of initiating only a moderate number of transactions with a limited amount of data changes per minute Thousands of end-users can already initiate thousands of transactions and generate high I/O rates with only low data rates End-users are directly affected by the application response time Peoples work time is expensive Excellent overall response time of the application is business critical and requires low I/O response times at high I/O rates
14 2011-04-04
Batch Jobs A single batch job can already generate a considerable amount of disk I/O operations in terms of I/O rate and data rate Multiple batchjobs can create a huge amount of disk activity Batch jobs should not interact with end-user transactions and are typically run outside end-user business hours Time frames for batch jobs even during nights / weekends are limited Overall job runtime is critical and mostly dependent on achieved overall data rate
2011 IBM Corporation
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
15
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Additional workload performance characteristics / objectives: Read cache hit ratio (percentage of read cache hits) average response time (RT) requirements (e.g. RT < 10ms)
16
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: 2.5" & Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
17
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Average Seek Time [ms] (head movement to required track) Rotational Latency [ms] (disk platter spinning until the first sector addressed passes under the r/w heads; avg. time = half a rotation) Transfer Time [ms] (read/write data sectors, 1 sector = 512 Byte)
Start
18 2011-04-04
Seek Time
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
= see manufacturer specs (typical: 4-10ms) = (60000/RPM) [ms] (typical: 2-4ms) = 1000 sectors sector size / avg. Transfer Rate [ms] (typically << 1ms for small I/O request sizes 32kB)
2011 IBM Corporation
19
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
This is just an example for getting a view on typical disk drive characteristics. The chosen disk types above do not necessarily represent the characteristics of the disk drive modules used in IBM System Storage systems. Source: www.seagate.com (2008)
20 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Rules of Thumb - Random IOps/HDD (conservative estimate to start with): FC 15k DDM : ~160 IOps FC 10k DDM : ~120 IOps SATA2 7.2k DDM: ~75 IOps
21 2011-04-04
A single disk drive is only capable of processing a limited number of I/O operations per second!
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Intelligent Cache Page Replacement & Prefetching Algorithms Standard: LRU (least recently used) / LFU (least frequently used) IBM System Storage DS8000 - Advanced Caching Algorithms 2004 ARC (Adaptive Replacement Cache) 2007 AMP (Adaptive Multi-stream Prefetching) 2009 IWC (Intelligent Write Caching)
IBM Almaden Research Center - Storage Systems Caching Technologies http://www.almaden.ibm.com/storagesystems/projects/arc/technologies/
22 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Tagged Command Queuing (TCQ, SCSI-2) & Native Command Queuing (NCQ, SATA2) further improves disk drive random access performance by re-ordering the I/O commands so that workloads can experience seek times which are considerably less than the nominal seek times Queue Depth: SATA2 (NCQ): 32 in-flight commands, SCSI-3 (TCQ): 2^64 in-flight commands
23 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
100% 100%
50% 50%
25% 25%
Disk Subsystem Cache Read Cache Hits Write Cache Hits / Write behind Sequential Prefetch Algorithms
1:500 1:1000 1:500 1:1000
Intelligent Cache Page Replacement & Prefetch Algorithms What data should be stored in cache based upon the recent access and frequency needs of the hosts (LRU/LFU)? Determine what data in cache can be removed to accommodate newer data. Predictive algorithms to anticipate data prior to a host request and loading it into cache.
26
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Even with reduced average seek times you cannot expect more than a few hundred random I/O operations per second from a single HDD. So a single HDD can only process a limited number of random IOps with average access times in the typical range of 5...15ms due to the mechanical delays associated with spinning disks (HDDs).
27 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Storage Disk Subsystem Typical I/O Rate & Response Time Relation
Response Time versus I/O Rate
30 25 20 15 10 5
28
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: 2.5" & Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
29
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Performance:
number and speed of disk drives (spindles) to meet IOps requirements high no. of fast, low capacity drives required to meet performance needs
Cost:
Performance
IOps IOps
no. of drives
Capacity
drive capacity
GB GB
higher
lower
COST
30 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
146GB15k drives are an excellent trade off between performance and capacity needs
2011 IBM Corporation
2005 0.7
IOps IOps = Access Density GB GB [IOps/GB]
cold data
2011 IBM Corporation
Access Density is a measure of I/O throughput per unit of usable storage capacity (backstore). The primary use of access density is to identify a range on a response time curves to give the typical response time expected by the average customer, based on the amount of total usable storage in their environment. The average industry value for access density in the year 2005 is thought to be approximately 0.7 I/Os per second per GB. Year-to-year industry data is incomplete, but the value has been decreasing as companies acquire usable storage faster than they access it.
31 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Subsystem Sizing Meeting Performance and Capacity Requirements Application: Capacity 1000GB; Performance 1000 IOps (1.0 IOps/GB)
SATA
Access Density: 1.1 IOps/GB
FC
SATA
Access Density: 0.075 IOps/GB
SATA
7x 146GB15k FC
(160 IOps/HDD; 15W)
1120 IOps
1022 GB 105 W
32 2011-04-04
75 IOps
1000 GB 9.8 W
1050 IOps
14000 GB (!) 137.2 W
2011 IBM Corporation
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
SATA-1
33 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
SATA 7.2k
34
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
35
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Parity
(2) (4)
37
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Parity
Cache = data being read from disk = data being written to disk
38
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
700 Reads 50% Cache Hits = 350 Reads 300 Writes 2 (two mirrored writes) = 600 Writes a total of 950 physical IOps on the disks at the physical backend
RAID10 already outperforms RAID5 in a typical 70-30-50 workload.
!!! Consider using RAID10 if random write percentage is higher than 35% !!!
39 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
+ +
o +
+ +
+ o
40
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
RAID6 - Overview
RAID6: Dual parity RAID
DS8000: 5+P+Q+S or 6+P+Q arrays (using modified EVENODD code) Survives 2 erasures 2 drive failures 1 drive failure plus a medium error, such as during rebuild (especially with large capacity drives) Like RAID5, parity is distributed in stripes, with the parity blocks in a different place in each stripe RAID6 does have a higher performance penalty on write operations than RAID5 due to the additional parity calculations.
Space efficiency
87.5%
50% 75%
2
6
2011 IBM Corporation
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
DS8000 R4.0 DS8000 R4.0 (no IWC) (no IWC) full stroke full stroke
42 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
RAID6 RAID6
RAID5 RAID5
RAID10 RAID10
DS8000 R4.0 DS8000 R4.0 (no IWC) (no IWC) full stroke full stroke
43 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: 2.5" & Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
44
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Example: DS8800 (2.5" 146GB15k SAS HDDs) vs. DS8700 (3.5" 146GB15k HDDs)
Fully configured with a base frame and two expansion frames, the new DS8800 can reduce floor space requirements by 40% and energy requirements by over 35%, all while supporting more drives than a five-frame DS8700 model (e.g. 1056x 2.5" disks in 3 frames in DS8800 vs. 1024x 3.5" DDMs in 5 frames DS8700). The small-form-factor drives offer better performance at the same rotational speeds, as well as better energy usage per drive and at a lower cost per gigabyte than the large-form-factor enterprise Fibre Channel drives available on most highend systems today. 16x 3.5" 24x 2.5" HDDs HDDs Estimated Storage Enclosure power DS8700 DS8800
Table takes into account controller card power, power efficiencies, power for cooling, and power for disks. Power per Enclosure Power per Disk 310 W 19.4 W 245 W 10.2 W
2011 IBM Corporation
45
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
=6.95W This is just an example for getting a view on typical disk drive characteristics. The chosen disk types above do not necessarily represent the characteristics of the disk drive modules used in IBM System Storage systems. Source: www.seagate.com (2010)
46 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
RAID6 RAID6
RAID5 RAID5
RAID10 RAID10
DS8700 R5.0 / / DS8800 R6.0 DS8700 R5.0 DS8800 R6.0 full stroke full stroke
47
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
1956 IBM RAMAC (1st disk drive) 5 MB storage, 1200 RPM data transfer rate 8800 characters per second 2010 Enterprise FC Hard Disk Drive (HDD) 600GB storage capacity, 15000 RPM data transfer rate 122 to 204 MB/s Last 50 years of HDD technology: HDD RPM: HDD Capacity 12.5 x 120 000 x
or ss ce ro P
Performance Gap
0.1 MHz
ce orman k Perf Di s
Time
48
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
RAID5 Write-Penalty RAID5 Write-Penalty (1:4 Backend Ops) (1:4 Backend Ops)
Single RAID5 Rank - Random Read Single RAID5 Rank - Sequential I/O SSDs show exceptionally low response times SSDs show exceptionally low response times Sequential I/O: SSDs ~~ HDDs Sequential I/O: SSDs HDDs
Source: IBM Whitepaper, IBM System Storage DS8000 with SSDs - An In-Depth Look at SSD Performance in the DS8000, http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101466
50 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
64kB Sequential IO
Source: IBM Whitepaper, IBM System Storage DS8700 Performance Whitepaper, ftp://public.dhe.ibm.com/common/ssi/sa/wh/n/tsw03053usen/TSW03053USEN.PDF 51 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
IOps (12x)
RT
52
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
IOps (>8x)
RT
53
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Solid State Drive (SSD) Latency and IOps vs I/O block size
Adding writes and/or increasing transfer size reduces SSD throughput and increases latency substantially
Source: Session sDS10, Storage Performance Made Easy with Easy Tier and SSDs, IBM, IBM STG Technical Conference, Lyon, 2010 54 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
CPU
1 cycle := 1 second
MEMORY
1:40 minutes
(<1ms) >100x
more IOps than HDD
SSD
11 days
56
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
cold
Solid State Drive technology remains more expensive than traditional spinning disks, so the two technologies will coexist in hybrid configurations for several years. Tiered storage is an approach of utilizing different types of storage throughout the storage infrastructure. Using the right mix of tier 0, 1, and 2 drives will provide optimal performance at the minimum cost, power, cooling and space usage. Data Placement is key! To maximize the benefit of SSDs it is important to analyze application workloads and only place data which requires high access densities (IOps/GB) and low response times on them.
IBM System Storage DS8000 with SSDs - An In-Depth Look at SSD Performance in the DS8000 http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101466 Driving Business Value on Power Systems with Solid State Drives ftp://ftp.software.ibm.com/common/ssi/sa/wh/n/pow03025usen/POW03025USEN.PDF
57 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
data
SSD whitepapers
Easy Tier optimizes SSD deployments by balancing performance AND cost requirements
Easy Tier delivers the full promise of SSD performance while balancing the costs associated with over provisioning this expensive resource
LUN Heatmap
Slower, inexpensive
Just Right
Fast, expensive
60
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
3X!
Easy Tier
T h r o u g h p u t (I O / s )
0:00
System configuration: 16x SSD + 96x 1TB SATA
2:00
4:00
6:00
8:00
10:00 Time
12:00
14:00
16:00
18:00
Source: Storage Performance Council, April 2010: http://www.storageperformance.org/results/benchmark_results_spc1#a00092 IBM Whitepaper, May 2010: IBM System Storage DS8700 Performance with Easy Tier, http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101675 61 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
96 SATA + 16 SSD
Single Frame
96 SATA HDD RAID10 plus 16 SSD RAID5
DS8700 R5.1
10000
20000
30000
40000
50000
60000
Throughput (IO/s)
62 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Smart data placement with Easy Tier: SPC-1 Backend I/O Migration
63
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
7.0
de ns ity )
2.0
us
1.0
en eo og
ac ce ss
heavily skewed medium skewed lightly skewed to predict the amount of the I/O workload that can be serviced by Solid State Drives (SSDs).
(h o
Sk
ew
Le
no
ve l
Skew Heavy
Medium Light
64
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
65
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
66
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
(1) Workload isolation (2) Workload resource-sharing (3) Workload spreading (1) Workload isolation (2) Workload resource-sharing (3) Workload spreading
Workload isolation (e.g. on extent pool and array level) dedicate a subset of hardware resources to a high priority workload in order to reduce impacts of less important workloads (protect the loved ones) and meet given service level agreements (SLAs) limit low priority workloads which tend to fully utilize given resources to only a subset of hardware resources in order to avoid impacting other more important workloads (isolate the badly behaving ones) provides guaranteed availability of the dedicated hardware resources but also limits the isolated workload to only a subset of the total subsystem resources and overall subsystem performance Workload resource sharing multiple workloads share a common set of subsystem hardware resources, such as arrays, adapters, ports single workloads now can utilize more subsystem resources and experience a higher performance than with only a smaller subset of dedicated resources if the workloads do not show contention with each other good approach when workload information is not available, with workloads that do not try to consume all the hardware resources available, or with workloads that show workload peaks at different times Workload spreading most important principle of performance optimization, applies to both isolated workloads and resource-sharing workloads simply means using all available resources of the storage subsystem in a balanced manner by spreading the workload evenly across all available resources that are dedicated to that workload, e.g. arrays, controllers, disk adapters, host adapters, host ports host-level striping and multi-pathing software may further help to spread workloads evenly
67 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
2.5 (SFF) SAS, 6Gbps SAS to disks Supports 24 disks per enclosure
70
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Agenda
Disk Storage System Selection & Specs Application I/O & Workload Characteristics Hard Disk Drive (HDD) Basics Its all mechanic HDD Performance & Capacity Aspects (SATA vs FC/SAS) RAID Level Considerations (RAID-5 / RAID-6 / RAID-10) New Trends & Directions: Solid State Drive (SSD) Basic Principles for Planning Logical Configurations Performance Data Collection and Analysis
71
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
(b) Storage Subsystem Performance Data Collection: DS3k/DS4k/DS5k (SMcli), XIV (XCLI), DS6k/DS8k and other (TPC for Disk)
72 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
only counters for quantity of processed I/Os up to current point in time no counters for quality of processed I/Os as, for example, I/O service times additional host system performance statistics required for I/O response times
73 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Always collect the Performance Statistics together with latest Subsystem Profile to document the actual subsystem configuration used during data collection
75 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
78
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
MS Excel spreadsheet for aa MS Excel spreadsheet for quick import and analysis quick import and analysis of DS4000 performance of DS4000 performance statistic outputs and statistic outputs and DS4000 profile with DS4000 profile with export feature for export feature for generating aahtml report generating html report
http://w3-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/PRS3088
79
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
80
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
81
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
IBM.2107-7503461
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
84
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
85
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
86
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
87
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
88
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
89
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
DS4000 and other supported SMI-S compliant storage subsystems: By Storage Subsystem Dont forget to export a complete set of reports for the subsystem of interest, e.g. for a DS8000: By Volume 20080131-75APNK1-subsystem.csv, By Port
20080131-75APNK1-controller.csv,
Some reports may give more or less data, depending on the exact level of SMI-S compliance by the vendor supplied CIM agents.
90 2011-04-04
Limit the reports to a representative time frame as the amount of data especially for the volume report can be extremly large!
2011 IBM Corporation
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
91
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
- small transfer sizes (4kB...16kB) with high I/O rates - low front-end response times around 5ms commonly expected
Backup, batch or sequential-like workloads
- large transfer sizes (32kB...256kB) with low I/O rates but high data rates - high front-end response times even up to 30ms still can be acceptable
Subsystem level front-end metrics (subsystem total average):
- Back-end Read Response Time < 25ms - Disk Utilization Percentage << 80% - I/O rate: depends on RAID level, workload profile, number and speed of DDMs
humb s of t le as ru s tions ges shold t e sug e thre riat nmen t som access): nviro pprop e jus r la ar e lues a enera rticul se va . In g s pa h nt The rt wit nts. e clie to sta ireme on th d qu base ion re to be plicat p need and a
considered very busy with I/O rates near or above 1000 I/Os (DS8000/DS6000)
Volume level front-end metrics (I/O performance as experienced by the host systems):
- Overall Response Time < 15ms (depends on application requirements and workload) - Write-cache Delay Percentage < 3% (typically should be 0%)
93 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
94
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
http://escc.mainz.de.ibm.com
95 2011-04-04 IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague 2011 IBM Corporation
Disclaimer
Copyright 2011 by International Business Machines Corporation. No part of this document may be reproduced or transmitted in any form without written permission from IBM Corporation. Product data has been reviewed for accuracy as of the date of initial publication. Product data is subject to change without notice. This information could include technical inaccuracies or typographical errors. IBM may make improvements and/or changes in the product(s) and/or programs(s) at any time without notice. Any statements regarding IBM's future direction and intent are subject to change or withdrawal without notice, and represent goals and objectives only. References in this document to IBM products, programs, or services does not imply that IBM intends to make such products, programs or services available in all countries in which IBM operates or does business. Any reference to an IBM Program Product in this document is not intended to state or imply that only that program product may be used. Any functionally equivalent program, that does not infringe IBM's intellectually property rights, may be used instead. It is the user's responsibility to evaluate and verify the operation of any on-IBM product, program or service. The performance information contained in this document was derived under specific operating and environmental conditions. The results obtained by any party implementing the products and/or services described in this document will depend on a number of factors specific to such partys operating environment and may vary significantly. IBM makes no representation that these results can be expected in any implementation of such products and/or services. Accordingly, IBM does not provide any representations, assurances, guarantees, or warranties regarding performance. THE INFORMATION PROVIDED IN THIS DOCUMENT IS DISTRIBUTED "AS IS" WITHOUT ANY WARRANTY, EITHER EXPRESS OR IMPLIED. IBM EXPRESSLY DISCLAIMS ANY WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE OR NONINFRINGEMENT. IBM shall have no responsibility to update this information. IBM products are warranted according to the terms and conditions of the agreements (e.g., IBM Customer Agreement, Statement of Limited Warranty, International Program License Agreement, etc.) under which they are provided. IBM is not responsible for the performance or interoperability of any non-IBM products discussed herein. The provision of the information contained herein is not intended to, and does not, grant any right or license under any IBM patents or copyrights. Inquiries regarding patent or copyright licenses should be made, in writing, to: IBM Director of Licensing, IBM Corporation, North Castle Drive, Armonk, NY, 10504-1785, U.S.A.
96
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague
Trademarks
The following are trademarks of the International Business Machines Corporation in the United States, other countries, or both.
AS/400, e business(logo), eServer, FICON, IBM, IBM (logo), iSeries, OS/390, pSeries, RS/6000, S/30, VM/ESA, VSE/ESA, WebSphere, xSeries, z/OS, zSeries, z/VM, System i, System i5, System p, System p5, System x, System z, System z9, BladeCenter, System Storage, System Storage DS, TotalStorage For a complete list of IBM Trademarks, see www.ibm.com/legal/copytrade.shtml Not all common law marks used by IBM are listed on this page. Failure of a mark to appear does not mean that IBM does not use the mark nor does it mean that the product is not actively marketed or is not significant within its relevant market. Those trademarks followed by are registered trademarks of IBM in the United States; all others are trademarks or common law marks of IBM in the United States.
97
2011-04-04
IBM Systems Technical University & STG Technical Enablement Conference, April 2011, Prague