ssd testing demystified for ssda - · pdf filehdd zone map ssd ftl map fixed zone map map on...
TRANSCRIPT
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 1
SSD Testing Demystified For SSDA Joseph Chen, ULINK Technology, Inc.
May 2nd, 2013
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 2
Agenda
1. SSD Testing Overview
• What is the Difference of HDD and SSD Specification?
• How Do You Test the SSD Specification?
• What do you test for?
2. SSD Test Data Review
• SATA-IO Digital Test Results
• Protocol Test Results
• Regression Test Results
3. FTL Testing • What is FTL Testing?
• How Do You Test FTL?
4. Endurance Testing • The TBW Mystery
• WAF Factor
5. Performance, Really?
• Real World Benchmark?
• TRIM Your Performance?
6. Summary • Magic or Myth?
• Ad Hoc SSD Testing
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 3
Using SSD as Primary System Storage
3
Wait!
One More Step…
Coexist and/or Replace HDD
• A huge market is waiting
• On Par w/ HDD Function and Reliability
System Cache Device
• Bridge Between DRAM and external storage
• Compatible with DRAM and HDD function and reliability
Premium Price/Performance Solution
• Can we deliver premiere quality for the price?
• Is an all-electrical components set better?
• What is the MTBF, AFR?
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 4
SSD Product Requirement Matrix
Reliability
• MTBF > 1M Hours
• UBER 10-15
• Endurance Rating
Performance
• How fast is the 4KB random IOPS?
• How does the TRIM improve performance?
Functional
• Does the SSD conform to SATA & ATA spec?
• Is the SSD FW and HW designed to spec?
• Does the SSD work on various systems?
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 5
HDD & SSD Spec Comparison HDD SSD Comment
General Form Factor 2.5” 2.5" Interface SATA 6Gb/s SATA 6Gb/s Power (Typ/Idle) 1.4W/590mW 6W/650mW NAND Class MLC, 25nm SSD Capacity 500GB/750GB/1TB 100GB/200GB/400GB
RPM 5400 HDD Performance Seq R/W (MB/s) 144 500/460 Random 4K R/W (IOPS) 75K/35K SSD
POR Time (Ave Sec) 8S 2S Cache 8MB HDD
Reliability UBER 10-15 10-15
MTBF (Hours) 1M 2M Endurance TBW SSD
Load/Unload Cycles 60K HDD Function Standard SATA 3.0, ATA-ACS2 SATA 3.0, ATA-ACS2 Features SMART, NCQ, SMART, NCQ, Trim, AES-256 Power Loss Protection Super Cap SSD Environment Shock/Vibration (Gs) 400 1000
Temperature (C) 0-60 0-70 Acoustics (dBA) 24 HDD
Compliance Logo SATA-IO/WHQL SATA-IO/WHQL Agency UL,CE,RoHS UL,CE,RoHS
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 6
SSD Testing Overview EVT/DVT RDT
Function Test
Protocol/Command Test Electrical/Power Test EMI/Agency/Logo Test Environmental Test
Reliability Test
Data Compare Test POR/Power Cycle Error Handling Test Mini RDT
MTBF Test Endurance Test UBER Test
Performance Test
IOPS Test MB/s Test Command Service Time Test TRIM Performance Test
Compatibility Test
System Install/Boot Test Application SW Test System Operation Test
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 7
SSD Validation Process
DVT (2-4 Weeks)
•Function
•Reliability
•Performance
DVT 2… (2-4 Weeks)
•Function
•Reliability
•Performance
Compatibility Test
•Function
•Reliability
•Performance
Environmental Test
•Electrical
•Environmental
•Agency
RDT (6 Weeks)
•MTBF
•Endurance
•UBER
RTS Ship It!
DVT for SSD Controller Vendors
DVT for SSD Module Manufactures
DVT for OEM SSD Qualification
DVT is the base for all SSD tests
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 8
SATA-IO Device Digital Test Summary DUT Sum SDG1 SDG2 SDG3 SDG4 SDG5 SDG6
Errors GTR NCQ ASR SSP IPM DOF H1 0
H2 0
H3 1 1
H4 1 1 H5 0
H6 1 1
H7 0 H8 0
S1 5 1 4
S2 0
S3 1 1
S4 0
S5 0
S6 0
S7 0 S8 0
S9 4 1 2 1
S10 4 1 2 1
S11 0
S12 1 1
S13 0
S14 0
S15 0
S16 0
S17 0
S18 1 1
Test Abb Name SDG1 GTR General Test Requirements SDG2 NCQ Native Command Queuing SDG3 ASR Asynchronous Signal Recovery SDG4 SSP Software Settings Preservation SDG5 IPM SATA Interface Power Management SDG6 DOF Digital Optional Features
1. Total Fail Counts: 19 2. Top Fail Item: NCQ: 6 3. Total Fail Rate: 34.6% (9/26) 4. HDD Fail Rate: 37.5% (3/8) 5. SSD Fail Rate: 33.3% (6/18)
HDD SSD
Good 5 12
Bad 3 6
Ugly 0 0
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 9
Protocol Test Summary Test Abb Name PTC1 SEN SecurityEraseNormal PTC2 IDF IdfyInfo_SATA PTC3 SFS SetFeature_SATA PTC4 MAN MandatoryCmds PTC5 WCF WrCacheFlushTime PTC6 RWB RdWrBoundaryCk PTC7 ATA ATACmds PTC8 DID DCO_IdfyInfo PTC9 DCS DCO_SATA PTC10 SMS SmartSet PTC11 SCT SCT PTC12 PWM PwrMgt PTC13 PWS PowerState PTC14 IPM IPM_Cmplt PTC15 SST SSPState PTC16 SSP SSP PTC17 PHY PhyEvntCnt PTC18 DSM ATA_DSM PTC19 NCQ NCQ PTC20 SEC SecuritySet
DUT Sum PCT1 PCT2 PCT3 PCT4 PCT5 PCT6 PCT7 PCT8 PCT9 PCT10 PCT11 PCT12 PCT13 PCT14 PCT15 PCT16 PCT17 PCT18 PCT19 PCT20
Errors SEN IDF SFS MAN WCF RWB ATA DID DCS SMS SCT PWM PWS IPM SST SSP PHY DSM NCQ SEC H1 5 NS 5
H2 0 NS
H3 10 1 3 1 NS 5
H4 10 1 3 1 NS 5 H5 7 3 3 . NS 1
H6 3 . NS
H7 3 NS H8 1 NS
S1 55 6 3 NS NS NS 4 12 NS NS NS NS 6 5 19
S2 22 1 3 3 3 1 11
S3 3 1 2 NS NS
S4 BRK BRK NT
S5 7 2 5
S6 7 2 5
S7 18 NS 3 3 NS 7 5 NS S8 18 NS 3 3 NS 7 5 NS
S9 5 NS NS 5
S10 5 NS NS 5
S11 7 NS NS 2 5
S12 7 NS NS 2 5
S13 2 NS NS 2
S14 2 NS NS 2
S15 13 1 3 7
S16 13 1 3 7
S17 12 1 3 7
S18 5 1 1 1 2
0/26 0/26 0/26 0/26 0/26 1/26 10/26 1/18 2/18 4/25 9/23 3/25 1/25 0/22 0/24 2/24 0/24 7/25 17/24 5/23
1. Total Fail Rate: 14% (64/480) 2. Top Fail Tests: NCQ/SCT/ATA/SEC 3. Average Fail Items: 2.4 (64/26) 4. Average HDD Fail Count: 4.8 (39/8) 5. Average SSD Fail Count: 11.1 (201/18) 6. Broken: 1 SSD
Legend NT Not Tested BRK Broken NS Not Supported
HDD SSD
Good 4 0
Bad 4 17
Ugly 0 1
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 10
Regression Test Summary DUT Sum REG1 REG2 REG3 REG4 REG5 REG6 REG7 REG8 REG9 REG10 REG11 REG12 REG13 REG14 REG15
Errors SEN SEE STT DSN SWN RDC RPT RDT PWS IPM DCP MD5 SMT PIT PIW H1 3 1 2 H2 1 1 H3 10059 5 2 H4 11954 5 5 NS 11811 4 121 6 2 H5 10101 2 9 21 H6 5 1 1 2 1 H7 14849 2 7 1 H8 10011 1 1 2 10000 4 3 S1 1 NS NS NS NS 1 NS BRK NT S2 0 2 2 BRK NT S3 1323 NS NS NS BRK S4 0 NT/BRK S5 9 1 9 BRK S6 0 1 BRK NT S7 106 NS NS 1 1 NS 4 100 S8 5 NS NS 1 NS 4 S9 2 1 1 BRK NT
S10 2 1 1 BRK NT S11 2490 1 NS HANG 20 2470 NT BRK S12 0 NS BRK NT S13 10 10 NS BRK NT S14 11 1 10 NS BRK NT S15 10 6 2 3 6 1 NT S16 11 5 2 3 3 1 NT S17 13 4 2 6 1 NT S18 4 4 BRK NT
4/23 4/22 12/23 11/24 2/18 4/23 1/22 2/22 6/22 1/18 8/18 4/17 6/16 6/15 7/13
Test Abb Name REG1 SEN SecurityEraseNormal REG2 SEE SecurityEraseEnhanced REG3 STT SctTemperatureTest REG4 DS1 DstShortOfflineTiming REG4 DS2 DstShortCaptiveTiming REG4 DS3 DstLongOfflineTiming REG4 DS4 DstLongCaptiveTiming REG4 DS5 DstConvOfflineTiming REG4 DS6 DstConvCaptiveTiming REG4 DS7 DstSelOfflineTiming_SmallSpan REG4 DS8 DstSelCaptiveTiming_SmallSpan REG4 DS9 DstSelOfflineTiming_HalfDisk REG4 DS10 DstSelCaptiveTiming_HalfDisk REG5 SW1 SctWrSameTime_PartialBeginDisk REG5 SW2 SctWrSameTime_PartialLastDisk REG5 SW3 SctWrSameTime_WholeDisk REG6 RDC RandomCmds REG7 RPT RandomPIOCmdTime_Distribution REG8 RDT RandomDMACmdTime_Distribution REG9 PWS PowerState REG10 IPM IPM_Cmplt REG11 DCP DataCmp_NCQ_DMA_PIO REG12 MD5 MD5 REG13 SMT SctMaxTemperatureTest REG14 PIT PwrInterrupt REG15 PIW PwrInttrruptWDataComp
1. Total Fail Rate: 26% (78/296) 2. Top Fail Items: PIW/DCP 54%/44% 3. Total Broken Rate: 48% (12/25) 4. Top Broken Items: PWS/PIW/PI/STT 5. HDD Broken Rate: 0% (0/8) 6. SSD Broken Rate: 70% (12/17)
Legend NT Not Tested BRK Broken NS Not Supported HDD SSD
Good 7 2
Bad 1 3
Ugly 0 12
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 11
Myth on Reliability Test
Should SSDs and HDDs be tested the same way?
Test Spec HDD SSD
MTBF 1M Hours Mechanical - Spindle
Head/Media
NA
Load/Unload 60K Mechanical - Spindle
Head Stiction
NA
Power Cycle 50K Spindle Motor Ramp/Latch/Park
NA
Endurance TBW NA NAND
Power Interrupt 5K NA FTL
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 12
FTL in the SSD FW Block Diagram
Host Cmds
Wear Leveling
NAND
Garbage Collection
Command Decode
Cache/Queue
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 13
FTL LBA Mapping Comparison
HDD Zone Map SSD FTL Map
Fixed Zone Map Map on cylinder/head/sector Map created on factory Burn-in
Variable Mapping High degree of variance Complicate meta data management
Static Mapping Map does not change on host command
Dynamic Mapping Map changed on host command
No Update Map does not change on Idle
Continuous Update Mapping Map changed on background tasks (Wear Leveling/Garbage Collection)
Slow Mapping Mechanical seek/latency (~200 IOPS, 5ms)
Fast/Ultra Fast Mapping High IOPS (~50K, 20us) 0.4% background mapping
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 14
How to Test FTL
FTL Stability
Stress Test
• Speed Stress
• Access Stress
• S1/S3 Power State Stress
Error Handle Test
• Protocol Error
• Interface Error
Aging Test
• Growing Defects
• Read Disturbance
• Program Disturbance
Disruptive Test
• Power Disruption
• Voltage Disruption
• Asynchronous Events
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 15
Myth on Endurance Test Variations of the SSD Endurance Specification Varies? Name Descriptions Workload
Specified
DWPD (Drive Write Per Day)
• Ten full drive write per day, 5 year warranty (Seagate) • Ten drive write per day for 5 years (Intel/Seagate)
No
GBPD (Giga Byte Per Day)
• Minimum of three years of useful life under typical client 3 years workloads with up to 20 GB of host writes per day (Intel)
• 20GB/day of host writes for 5 years under typical client workloads (OCZ)
Typical Client Workload?
TBW (Tera Byte Write)
• TBW: Total bytes written (Micron/Kingston/WD) • TBW: Tera byte write (Sandisk)
No
PBW (Peta Byte Write)
• Lifetime endurance (8KB): Up to 14 PB (Intel) • 4 kilobytes (KB) write endurance of up to 1.5 petabytes (PB) with 20
percent over-provisioning (Intel)
Yes, 4KB/8KB
JEDEC 218/219
• JEDEC TBW: JEDEC 218/219 TBW rating with client/enterprise workload Yes
Note: Information quoted from the public specifications of the list companies for the purpose of illustration
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 16
Endurance Test Workload Is 4K/8K Random Write Suitable for Endurance Tests?
Issue: TBW Definition
Not real world workload
TBW is based on WAF, WAF is based Workload
Solution: Use JEDEC Workload for Endurance Tests
𝑻𝑩𝑾 =(𝑺𝑺𝑫 𝑪𝒂𝒑𝒂𝒄𝒊𝒕𝒚 ∗ 𝑵𝑨𝑵𝑫 𝑪𝒚𝒄𝒍𝒆𝒔) ∗ (𝟏 + 𝑶𝑷)
𝟐 ∗ 𝑾𝑨𝑭 𝑾𝒌𝒍𝒅
𝑻𝑩𝑾 = 𝑻𝒆𝒓𝒂 𝑩𝒚𝒕𝒆 𝑾𝒓𝒊𝒕𝒆 𝑬𝒏𝒅𝒖𝒓𝒂𝒏𝒄𝒆 𝑹𝒂𝒕𝒊𝒏𝒈
𝑶𝑷 = 𝑶𝒗𝒆𝒓𝒑𝒓𝒐𝒗𝒊𝒔𝒊𝒐𝒏 = 𝑷𝒉𝒚𝒔𝒊𝒄𝒂𝒍 𝑪𝒂𝒑𝒂𝒄𝒊𝒕𝒚
𝑳𝒐𝒈𝒊𝒄𝒂𝒍 𝑪𝒂𝒑𝒂𝒄𝒊𝒕𝒚− 𝟏
𝑾𝑨𝑭 = 𝑾𝒓𝒊𝒕𝒆 𝑨𝒎𝒑𝒍𝒊𝒇𝒊𝒄𝒂𝒕𝒊𝒐𝒏 𝑭𝒂𝒄𝒕𝒐𝒓 𝟐 = 𝑮𝒖𝒂𝒓𝒅 𝑩𝒂𝒏𝒅 𝒇𝒐𝒓 𝒘𝒆𝒂𝒓 𝒍𝒆𝒗𝒆𝒍𝒊𝒏𝒈 𝒆𝒇𝒇𝒆𝒄𝒕
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 17
Myth on Performance Test
Current SSD Benchmarked on 4K Random R/W
SNIA Standardized SSD Performance
Issues – “Unreal”
No real world workload
No TRIM performance
Solution – JEDEC 218 Client Workload
Real world workload
TRIM commands supported
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 18
Performance Test Summary
DUT Capacity IOPS
S1 128 GB 1186
S2 256 GB 1879
S3 100 GB 5241
S4 160 GB 2783
S5 80 GB 2442
S6 80 GB 813
S7 120 GB 1632
S8 128 GB 1997
S9 250 GB 1363
S10 250 GB 1050
S11 480 GB 3489
S12 240 GB 2023
S13 120 GB 1856
0
1000
2000
3000
4000
5000
6000
1 2 3 4 5 6 7 8 9 10 11 12 13
IOP
S
DUT
JEDEC Wkld Performance
JEDEC 218A Client Workload IOPS
Tested with JEDEC Client Workload with 38M Commands Issued
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 19
Trim Performance Comparison
0
1000
2000
3000
4000
5000
6000
1 2 3 4 5 6 7 8 9 10 11 12 13
IOP
S
DUT
Trim Performance Comparison
No TRIM TRIM
DUT GB IOPS no Trim
IOPS Trim % Diff
S1 128 GB 1186 772 65.09%
S2 256 GB 1879 667 35.50%
S3 100 GB 5241 4615 88.06%
S4 160 GB 2783 3368 121.02%
S5 80 GB 2442 2944 120.56%
S6 80 GB 813 1559 191.76%
S7 120 GB 1632 2430 148.90%
S8 128 GB 1997 2542 127.29%
S9 250 GB 1363 1185 86.94%
S10 250 GB 1050 1132 107.81%
S11 480 GB 3489 2490 71.37%
S12 240 GB 2023 1753 86.65%
S13 120 GB 1856 1644 88.58%
Tested with JEDEC Client Workload with 38M Commands Issued
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 20
Trim Performance 1 (Improved)
0
1000
2000
3000
4000
5000
6000
7000
1 51 101 151 201 251 301 351
IOP
S
IO Count (in 100K)
Trim Performance Comparison JEDEC 218 Client Wkld
No Trim Snapshot No Trim Average Trim Snapshot Trim Average
Tested with JEDEC Client Workload with 38M Commands Issued
Trim IOPS=1559
No Trim IOPS=813
Average IOPS
Instantaneous IOPS
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 21
Trim Performance 2 (Reduced)
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
1 51 101 151 201 251 301 351
IOP
S
IO Count (in 100K)
Trim Performance Comparison JEDEC 218 Client Wkld
No Trim Snapshot No Trim Average Trim Snapshot Trim Average
Tested with JEDEC Client Workload with 38M Commands Issued
Trim IOPS=2490
No Trim IOPS=3489
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 22
Summary
SSD Testing: Magic or Mystery?
Steps to Make SSD to a Premium Product Function
Protocol Test Regression Test for FTL
Reliability JEDEC Workload Endurance Test
Performance Benchmark JEDEC Real-Life Workload IOPS Benchmark TRIM Performance
The Ad Hoc SSD Testing Suites Fills the Gap!
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 23
THANK YOU!
ULINK Technology, Inc.
3120 De La Cruz Blvd, Ste 117
Santa Clara, CA 95054
1-408-446-8455
www.ulinktech.com
Contact (at) ulinktech.com
美商宇凌科技有限公司 台灣分公司 新竹縣竹北市高鐵七路65號16樓之17號
886-3-668-1770
5/2/2013 Copyrighted @ULINK Technology, Inc. Page 24