Desktop, Nearline & Enterprise Disk Drives What’s the difference? Willis Whittington, Seagate Technology Feb 21st, 2008 Abstract Desktop, Nearline & Enterprise Disk Drives What’s the difference? For the past twenty five years the storage marketplace has been divided into two major categories namely “Desktop” and “Enterprise”. Recently, a third player variously known as “Nearline”, “Reference” or “Business Critical” has evolved to provide a low cost, high capacity storage solution for Enterprise data that no longer needs to exist in a high availability transactional processing environment but must maintain 24 x 7 availability as a reference or backup resource. Each of these classes of drives requires a unique and specific set of attributes to fulfill its role. This presentation will explore these differences and explain why you need to use the right drive for the right application. Agenda Basic Comparisons SAS & SATA Compatibility The Advantages of Nearline SAS Rotational Vibration Data Error Rate Error Correction Capability Data Integrity Performance Annualized Failure Rate ~ Q & A along the way ~ Basic Comparisons Comparison Table DT / NL / MC* Key: Metric Good Better Best Desktop (DT) SATA Nearline (NL) SAS Nearline (NL) Enterprise MC* Capacity (GB) 1,000 1,000 1,000 450 Cost Low Mid Mid High 1x 1.2x 1.2x+ 1.5x 600,000 1,200,000 1,200,000 1,600,000 Low (<10%) Low/Medium (<20%) Low/Medium (<20%) High (100%) Parity (?) EDC + (ECC?) EDC/ECC + Proprietary Data Integrity Protection EDC/ECC + Proprietary Data Integrity Protection Unrec Error Rate 10-14 10-15 10-15 10-16 RV Radians/sec2 6 12.5 12.5 >21 SATA SATA + Time Control Full SCSI Full SCSI Firmware/Features Standard SATA SATA + Selected Nearline Features SCSI + Adv. Features SCSI + Adv. Features (Enabled by Dual CPU) (Enabled by Dual CPU) Power On Hrs/Year 2400 Power Consumption MTBF (Hrs) Duty Cycle Data Integrity Error Recovery 8,760 8,760 (Low Duty cycle) (Low Duty cycle) 8,760 Multi Initiator No No 16 Hosts & Dual Port 16 Hosts & Dual Port Performance 1x 1x 1x+ 1.4x / 2.5x (Seq / Rand) T10 Data Protection No No Yes Low Low High + Dual Port Scalability *Mission Critical 2x Duplex Yes High + Dual Port 2x Duplex Anatomy of an Enterprise Drive Motor Higher rpm than NL or DT Tighter specifications Less runout More expensive Anatomy of an Enterprise Drive Discs Four platter design Smaller diameter than NL/DT Full media certification Fully characterized Variable sector format Anatomy of an Enterprise Drive Head Stack Eight head design Low mass, high rigidity Voice coil designed for o optimal performance o 100% duty cycle Higher cost design Anatomy of an Enterprise Drive Environmental Control Module Humidity Control Chemical Absorbtion Multi-point filtration Windage Design Anatomy of an Enterprise Drive Misc Mechanical Powerful Voice Coil Motor Stiffer Covers Air Control Devices Faster Seeks High Servo Sample Rate Low RV susceptibility Anatomy of an Enterprise Drive Electronics Dual processors Multi host queuing Dual port Twice the memory of NL/DT High rpm control Command scheduling Superior error protection Superior error correction Smart servo algorithms Perform. optimization Data integrity checks Sequential h/w assist Con>> SAS & SATA Compatibility I/O Connectors for SAS & SATA Vcc Enclosure Vcc Drive Enclosure Vcc Drive Vcc ~3msec Current Limited Steady State Data Rx Tx 3.3V Voltage Pins 5V 12V LED/ Spin-up Key Gnd For SAS, the key-way is filled in and its flip side is used for the 2nd Port << Con Gnd This prevents a SAS drive from being plugged into a SATA cabinet slot. The Advantages of Nearline SAS NL SATA Compared to NL SAS Stepping up to SAS provides Mission Critical Compatibility NEARLINE SAS Dual Port Full EDC & ECC 100% Phy Compatible End-to-End Data Integrity Variable Sector Size Multiple Host Support Full SCSI Command Set Enterprise Command Queuing Concurrent Data Channels Full Duplex (Bidirectional) I/O NEARLINE SATA SAS/SATA NL Physical Differences SAS Electronics SAS Port “B” SATA Electronics Nearline Head/Disc Assy. SAS/SATA NL Differences Dual Port Two Data Channels 2 Concurrent Writes OR 2 Concurrent Reads OR 1 Write + 1 Read SAS CHANNEL 2 Full Duplex CHANNEL 1 SAS SATA Xmits In both directions at one time Half Duplex One Data Channel 1 Write OR 1 Read SATA Xmits In one direction at one time Multi Host Command Queuing SAS SAS EXPANDER EXPANDER SAS Drive Queue Supports 16 Hosts Interposer Interposer to handle Q’ing for 2nd host SATA Drive NCQ* supports a single Host *Native Command Queuing NL SAS/SATA Summary Both SATA and SAS Nearline drives are designed for use in Enterprise Mission Critical environments. SAS Nearline drives have additional advantages which are made possible by the Serial SCSI interface and enterprise electronics: • Full system interface compatibility at the protocol, physical (“phy”), and command level • Enterprise error recovery and performance optimization controls • Full data integrity protection both within the drive and at the system level with DIF support. Rotational Vibration RV Emitted by a Seeking Drive • PS drives are not designed for Neighboring Drive’s backplane (JBOD/SBOD) use Servo needs to and are not equipped to cope compensate for with the effects of RV externally induced RV RV is Proportional to Seek Current Current in Voice Coil Disc Movement Coil Current ‘Scope Picture, Seagate Prod. Dev. HDA subjected to rotational forces Rotational Vibration Impact on Performance* 6 12.5 21 rads/sec2 - 120 Enterprise Enterprise 80 Barracuda ES Enterprise 40 Nearline 60 Desktop % Degradation 100 Nearline Desktop 20 Desktop 0 0 10 20 30 40 RV Level (Rad/sec2) 50 60 *Source: STX Competitive Analysis. RV in 33 Different Cabinets 45 Rotational Vibration 40 2 rads/sec (rms) 11 Unacceptable 35 Acceptable for Enterprise Nearline 30 22 Desktop 12 5 25 Enterprise 21 Rad/sec2 20 15 SATA Nearline 12.5 Rad/sec2 10 SATA Desktop 6 Rad/sec2 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Cabinet Tested 20 21 22 23 24 25 26 27 28 29 30 31 32 33 More stringent RV spec. needed for SATA cabinets RV aggravated by system fans, random access and “bursty” workloads Data Error Rate UER* on High Capacity RAID Sets The UER for SATA desktop is 1 in 1014 bits transferred • 1014 bits = 12½ terabytes A 500 Gbyte drive has 1/25 x 1014 bits Rebuilding a SATA drive in a RAID 5 set of 5 drives means transferring 5/25 x 1014 bits = 1/5 of UER spec. • 20% probability of an Unrecoverable Error during the rebuild. Better odds would be available with RAID 1 or 6 • RAID 1 rebuilds from a single mirror drive • RAID 6 can tolerate a second error during the rebuild. Risks can be reduced with good error management • Intelligent rebuild (ignore unused capacity) • Background media scan (dynamic certification) *Unrecoverable Error Rate DT / NL / MC UER* 20% Probability of Unrecoverable Errors during RAID Rebuild Probability of UE 16% Desktop Drives UER = 10-14 12% 8% 4% Nearline UER = 10-15 2% 0.2% 3 4 5 # of 500 GB Drives in RAID Set *Unrecoverable Error Rate Enterprise UER = 10-16 Error Correction Capability Standard vs Reverse ECC (Write Command) + User ec randomized ec User User Data ECC Generator User User Data Randomizer RLL Encoder + randomized RLL Encoder Randomizer encoded encoded ec ec encoded ECC Generator Standard vs Reverse ECC (Read Command) _ ec ?? ECC Correction De-randomize ec 120 Bit Error (Propagates in Decoder) 20 Bit Error RLL Decoder User Data _ De-randomize User User Data 20 Bit Error ec No Error Propagation RLL Decoder ec ECC Correction Error Corrected on the fly Sync Mark Errors on SATA Drives Sync Mark Sync Field Media Flaw Sector Format User Data Field ECC The Sync Field is used to get the read channel in frequency sync with the data recorded on the media The Sync Mark is used to define the beginning of the User Data Field Failure to recognize the Sync Mark (due to a thermal asperity or a grown media defect) means the User Data Field is not delineated and the data is lost. Sync Mark Errors on SAS Drives Sync Mark Sync Field Media Flaw n bytes Sync Mark 2 embedded in the data field User Data Field Sector Format ECC Read Operation ????? !!! Read Channel realizes the first Sync Mark is missing: - Loads Buffer with n zeroes - Starts searching for SM2 0011011000011110101011000011010100110110001010101000111 ECC Read Channel finds SM2 and reads the data following it into the Buffer, starting at location n+1 DATA BUFFER Before ECC 00000000000000 0011011000011110101011000011010100110110001010101000111 The missing n bytes are recovered using the ECC After ECC 11010111010011 0011011000011110101011000011010100110110001010101000111 Data Integrity Single Sector Data Transfer from the Host SOF User Data Header DIF CRC EOF FC or SAS Interface Interface Disparity Check Odd Parity Generator = CRC Generator NO = NO ERROR Drive I/O Checks DIF Generator Frame Buffer CRC DIF DIF Parity User Data path = NO User Data Odd Parity Generator Encryption Engine IOEDC/IOECC Engine IOEDC Generator ERROR NO Cache Buffer User Data DIF IOEDC EDC/ECC IOEDC/IOECC Engine RLL Encoder User Data DIF IOEDC DIF IOEDC ECC Reverse ECC Generator User Data = Randomizer YES User Data DIF IOEDC Performance Performance Comparison 100 Enterprise Sequential Access 90 80 Desktop / NL Sequential Access Mbytes/sec 70 Vendor Range 60 50 Sequential Transfer Rate α RPM x Disc Dia x Bit Density 40 Enterprise Transfer Rate Desktop Transfer Rate 30 = = 20 RPM x Disc Dia x Bit Density RPM x Disc Dia x Bit Density 15000 x 65 7200 x 95 = 142% (independent of seek time & Latency) 10 1 4 Q Depth 16 32 Performance Comparison 100 400 90 360 80 320 o Rand e s i r rp Ente Desktop / NL Sequential 70 m 280 60 240 50 200 40 160 do Desktop / NL Ran 30 m 120 20 80 10 40 1 4 16 32 Operations/sec Mbytes/sec Enterprise Sequential SPC-1C Performance Comparison SPC-1C Benchmark 30 Response ResponseTime time (ms) (ms) 25 Desktop 20 Vendor & Model Variance Enterprise 15 Desktop/NL 10 Enterprise 5 OLTP 0 0 100 200 IO/s IO/s 300 400 • SPC-1C comprises I/O operations demonstrating small storage subsystem performance (1-16 drives) while performing the typical functions of a business critical application. 500 Database Email SPC-1C Workload AFR (Annualized Failure Rate) AFR vs Duty Cycle Typical Enterprise application, Power-On-Hours, and Temperature 1.2 Desktop Drives AFR (%) 1 Equivalent Enterprise Duty Cycles 0.8 2½ x AFR Enterprise Drives 0.6 0.4 0.2 SATA drives in Enterprise applications run hotter, at higher duty cycle, and for more Power-On-Hours than in desktop applications. 0 0.5 Low 1 2 3 5 7 Duty Cycle (%) 10 15 20 High In Conclusion….. Although technological advances, driven by Enterprise research, will be leveraged into SATA products, there will continue to be functional limitations imposed on these devices by the overriding metric of Low $/GB Storage.