ProtecTIER on IBM i May,2011 Bob French Dynamix Group, Inc. bobfrench@dynamixgroup.com ™ © 2011 IBM Corporation ® ™ Agenda – ProtecTIER on IBM i • Product Overview – What does it do? – Deduplication Algorithms – Product Family – ProtecTIER on IBM i – where it fits – ProtecTIER on IBM i – how it attaches – ProtecTIER on IBM i - sizing – Next Steps 2 © 2011 IBM Corporation ® ™ ProtecTIER Product Overview 3 © 2011 IBM Corporation ® ™ ProtecTIER Vision and Design Criteria 1. 2. 3. 4. 5. 6. Data-agnostic factoring of up to 25 times or more Unmatched performance up to 1000 MB/s or more Unequaled scalability: up to 1 PB physical data Enterprise-class data-integrity: Not hash-based Simple, non-disruptive deployment Supported in most hardware and software environments No other dedupe technology meets all these criteria! 4 © 2011 IBM Corporation ® What does ProtecTIER do? ™ ProtecTIER IP Replication IBM i TS3500 Optional duplication to physical tape Ohio Minimized bandwidth since data is de-dup’d before sending (at local or remote site) New York IBM i ProtecTIER Virtual Tapes C Disk What is DeDuplication? B A B A C A B A C Local Saves to Virtual Tape with De-dup B A B A C A C A B A B C B A A A B 5 © 2011 IBM Corporation ® ™ Hyperfactor Deduplication in Action New Data Stream HyperFactor Repository Memory Resident Index Disk Arrays SAN Switch TS7650G Existing Data “Filtered” data IBM i Servers 6 © 2011 IBM Corporation ® ™ Deduplication Algorithms - Types Content Aware Hash Based Hash Value • Assumes the best candidate to Text against is an object de-dup with similar attributes (eg file name, file type) • De-dup ratios are lower since even a tiny difference within the boundary causes the match to fail • Sepaton ProtecTIER HyperFactor Pointer • Hash table grows as more backup data is stored • As the repository fills, backup speeds decrease since it takes longer to check the bigger table • Eventually the Hash Table can’t fit in memory and backup speeds decrease seriously • Hence this algorithm is only suited to smaller devices: it doesn’t scale well • Data Domain, FalconStor • Very fast algorithm since it processes backup stream to find possible matches then offloads work to disk to confirm the match • Algorithm can index 1 PB of physical disk using a 4 GB index that fits in memory, hence no performance degradation as the repository grows • Hence scalable for large repositories 7 © 2011 IBM Corporation Production Customers Deployment Results ® ™ 8 © 2011 IBM Corporation ® ™ Deduplication Algorithms – Post vs Inline Post Processing Deduplication • Backups run first without de-dup • Separate de-dup algorithm runs thereafter • Requires extra disk space to hold the interim full-sized copy of the backup • Used when the de-dup algorithm is not fast enough to run inline Inline De-Duplication (eg HyperFactor) • De-dup runs as part of backup process • Uses less disk • Once save is done, the entire process is done • Only possible with a fast de-dup algorithm like ProtecTIER HyperFactor 9 © 2011 IBM Corporation ® ™ IBM TS7600 ProtecTIER® Deduplication Family New in July 2010 Highest Performance Largest Capacity Good Performance Better Performance Highly Scalable Larger Capacity Highest Performance High Performance Highest Performance Largest Capacity High Capacity Largest Capacity High Availability High Availability Flexible Storage Scalable Low cost Good Performance Entry Capacity Very Low cost Single Node Single Node Single Node Single Node Single Node Up to 85 MB/sec Up to 85 MB/sec 5.9 TB (5.5 TiB) useable 4.4 TB (4.0 TiB) useable Up to 100 MB/sec 7 TB (6.3 TiB) useable Up to 250 MB/sec 18 TB (15.8 TiB) useable Up to 500 MB/sec 36 TB (31.5 TiB) useable Single Node Active-Active Cluster Active-Active Cluster Up to 900 MB/sec Up to 1500 MB/sec Up to 500 MB/sec 1 PB useable 1 PB useable 36 TB (31.5 TiB) useable Nominal Space Available = “useable” space * HyperFactor Ratio 1 TB = decimal TB = 1,000,000,000,000 bytes or 1,000 GB (i.e. 10^12 bytes) 1 TiB = binary TB = 1,099,511,627,776 bytes or 1,024 GiB (i.e. 2^40 bytes) 11 © 2011 IBM Corporation ProtecTIER Details ™ ® LTO-2 & LTO-3 Emulation TS7650 Appliance Notes: TS7650 Gateway TS7610 Appliance (1) IBM i has a max of 32 drives in a virtual library attached to a given server, and 92 drives total in a virtual library (2) IBM i has a max of 4096 cartridge locations in each library (slots + drives + IO slots + grippers) TS7610 Appliance TS7650 Appliance TS7650 Gateway # nodes 1 1-2 1-2 Max Throughput 80 MB/sec 100, 250, 500 MB/sec Up to 1500 MB/sec with 2 nodes Repository (Physical) 4.4 or 5.9 TB 7, 18, 36 TB Up to 1 PB Max # Virtual Libraries 4 12 16 Max # Virtual Drives 64 256 256 8,192 128,000 500,000 4 if TS7610 is the hub 12 if TS7650 is the hub 12 12 (1) Max # Virtual Cartridges (2) Replication – Max Spokes per Hub 12 © 2011 IBM Corporation ® ™ Customer Profile for each Appliance Configuration Ideal Customer for 7TB ProtecTIER Appliance 1 TB or less incremental backups per day 1-3 TBs full backups each week Experiencing average data growth Needs a cost effective solution Ideal Customer for 18TB ProtecTIER Appliance 3 TBs or less incremental backups per day 3-6 TBs full backups each week Experiencing rapid data growth Needs good performance to meet backup window Ideal Customer for 36TB ProtecTIER Appliance 5 TBs or less incremental backups per day 5-12 TBs full backups each week Additional growth expected Meeting the Backup window is an issue - higher performance needed * Note: These general guidelines are based on the backup workload that best fits each appliance configuration Please use Capacity Planning Tool to accurately size a solution to meet customer’s specific requirements 13 © 2011 IBM Corporation ® ™ IBM’s LTO Technology Roadmap 14 © 2011 IBM Corporation ® ™ ProtecTIER For IBM i 15 © 2011 IBM Corporation ® ™ ProtecTIER on IBM i – Where does it fit? Our Niche – ProtecTIER IP Replication For customers who are moving their tapes offsite via truck today and would like a safer, more automated solution IBM i ProtecTIER ProtecTIER Our Niche – Tired of Handling Tapes? IBM i Our Niche – Tiny LPARs Note: If the customer already has an HA/DR solution that replicates his data to his remote site, then that will likely provide a more economical solution for remote tape: • IBM i Software-based Replication (eg iCluster, MIMIX, Visions, iTera, etc) • External Disk Copy Services • IBM i Geographic Mirroring (formerly Cross Site Mirorring or XSM) For customers where a tape cartridge is much bigger than needed ProtecTIER ... VIOS / NPIV 16 © 2011 IBM Corporation ® ™ Virtual Tape on IBM i – Questions to Ask Your Vendor Overall Speed and Single Stream Speed Virtual Tape Devices shine when they can run a large number of mediumspeed backup streams. IBM i customers often need a small number of very fast streams. Be sure to understand the single stream performance provided to make sure your Virtual Tape Device will meet your needs IBM i Single Stream performance depends on the VTL disk type/amount Backup Scheduling LPAR 11 pm 11:30 pm Midnite 12:30 am 1 am 20 160 200 200 200 IBM i 01 IBM i 02 ProtecTIER 40-90 MB/sec per stream 40-90 MB/sec per stream 40-90 MB/sec per stream 40-90 MB/sec per stream TS7650 ProtecTIER Full box Save capacity is 1500 MB/sec with 2 nodes IBM i 03 IBM i 04 Total MB/Sec Draw a Backup Gantt Chart to check the MB/sec and # streams at your peak Non-Infinite Resources Current Technology Physical Drives run at 60-280 MB/sec per stream (umix / largefile) Although virtual tape if flexible, remember the resources aren’t infinite17 © 2011 IBM Corporation ® ™ ProtecTIER on IBM i – Support and Testing Supported with: • IBM i V5R4 onwards • Any IBM i fibre card supported on your server • BRMS is strongly recommended • Tested with the same COMPREHENSIVE Test Buckets used for regular tape drives IBM ProtecTIER is the ONLY External Virtual Tape product that is tested and supported by IBM Rochester 18 © 2011 IBM Corporation ® ™ ProtecTIER Attachment to IBM i - Details Fibre cards that use an IOP (fc 2765, 5704, 5761) IBM i V5R4M0 onwards TS7650 ProtecTIER Code Levels Min for Local Backups: V2.2.3.0 Min for IP Replication: V2.3.0 Fibre cards that don’t use an IOP (fc 5749, 5774/5276, 5735/5273, 5708 FCoE + Blades fibre cards) IBM i V6R1M1 onwards with the following PTFs IBM i 6.1.1: MF49234 + pre-reqs IBM i 7.1.0: MF49235 + pre-reqs POWER6 or POWER7 system TS7650 ProtecTIER Code Level V2.4.1.0 Server Code V2.4.3.0 PT Manager Code (GUI) BRMS is recommended since TS7650 presents as a tape library 19 © 2011 IBM Corporation IBM i IOPless Support for ProtecTIER Restrictions ® ™ Restriction #1: IBM i alt-IPL (reload) Restriction #2: TS7650 IPL with VIOS (this only applies to IOPless fibre cards, not the older IOP’d cards) VIOS IBM i IBM i SAN Switch TS7650 Node 0 Virt Drive 0 SAVSYS Tape Node 1 Virt Drive 2 Virt Drive 3 Virtual Library To D-IPL your IBM i, use TS7650 LUN masking so the adapter card can only see a single virtual drive (the one with the SAVSYS in it) TS7650 Other Tape in VIOS Zone If TS7650 is attached to VIOS, remove TS7650 port(s) from the VIOS SAN Zone before IPLing the TS7650, otherwise it may disrupt other devices 21 © 2011 IBM Corporation ® ™ BRMS DUPMEDBRM Compaction PTF TS7650 Virtual Tape Saves are not compacted so take 3x as much virtual media (gained back with dedup) Part of June 2010 BRMS PTF V5R4: SI38733 IBM i 6.1: SI38739 IBM i 7.1: SI38740 IBM i With the PTF, DUPMEDBRM can request compaction so uses less media TS3500 With PTF Before the PTF, dups used the same compaction parameter as the source volume, so more physical media was needed Exposes the COMPACT parameter so you can compact the physical volumes when you dup from ProtecTIER Before PTF Behavior: V5R4: control via Data Area Q1ADUPCOMP in QTEMP can be set to *FROMFILE, *YES, *NO IBM i 6.1 / 7.1 COMPACT(*YES) is available help text via web For new IBM i 6.1 auto-dup feature, need to change command default on DUPMEDBRM to *DEV Future releases: COMPACT(*YES) will be available with regular help text 22 © 2011 IBM Corporation ® ™ Sizing ProtecTIER For IBM i 23 © 2011 IBM Corporation ® ™ ProtecTIER on IBM i – Designing / Sizing Get the ProtecTIER on IBM i Introduction and Questionnaire IBMers: Partners: http://w3-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101536 http://partners.boulder.ibm.com/src/atsmastr.nsf/WebIndex/WP101536 Build a Repository Sizing Spreadsheet Complex Environment: Several days of work Build a Backup Schedule Gantt Chart (to figure out the peak MB/sec) LPAR GB in Save Iterations Kept GB in repository LPAR IBM i 01 200 GB 3 600 IBM i 01 IBM i 02 350 GB 7 2450 IBM i 02 IBM i 03 100 GB 3 300 IBM i 03 IBM i 04 575 GB 12 6900 IBM i 04 10250 Total MB/Sec Total Simple Environment: 1-2 hours of work 11 pm 11:3 0 pm 60 20 12: 30 am 1 am 60 60 60 80 80 80 80 60 60 200 200 20 80 20 mid nite 160 200 Then ask the ProtecTIER FTSS to tell you how many disk arms you need 24 © 2011 IBM Corporation ® ™ Next Steps 25 © 2011 IBM Corporation ® ™ If you would like to consider ProtecTIER for your shop … Detailed ProtecTIER Presentation Backup Environment Review Attend the ProtecTIER Hands-on Workshop Ask your ProtecTIER team to engage an IBM i / ProtecTIER specialist to review your backup environment with you • Two-Day Hands-on Workshop in Gaithersburg, Maryland. • Runs 1-2 times per month • No charge to attend Text Invite your local IBM ProtecTIER Sales Team to give you a moredetailed presentation 26 © 2011 IBM Corporation ® ™ Questions? 27 © 2011 IBM Corporation ® ™ Deduplication Market at a Glance DD880 DEDUP TECHNOLOGY ProtecTIER with HyperFactor RockSoft Hash-based Byte-level diff ! Potential Hash comparison collision Inline Deduplication Block Level DXi7500 RockSoft Hash-based VTL 700 SIR Hash-based ! Potential Hash ! Potential Hash collision collision Inline ! Post process ! Post process Block Level Block Level Block Level Deduplication PERFORMANCE Single node performance 500 MB/s ! 300 MB/s Dual node Cluster performance 1000MB/s !Clustering not ! 130 MB/s !Clustering not available available RESOURCE UTILIZATION No disk staging No disk staging area required area required Only 4GB RAM needed for a 1PB repository ! 188 MB/s S2100-ES2 DeltaStor Byte-level diff comparison See Note (3) Ø File Level See Note (4) ! 160 MB/s !Clustering with !Clustering with Global Dedupe not Global Dedupe not available available ! Staging area > Ø Staging area > than the size of largest full backup than the size of largest full backup twice the size of !Over 300GBs of RAM! of RAM! !Over 300GBs of RAM! See Note (2) ! Post process ! Staging area > !Over 300GBs See Note (1) See Notes (5-6) See Notes (7-8) See Notes (9-10) largest full backup 24GB of RAM See Note (11) Not hash based 28 © 2011 IBM Corporation ® ™ Deduplication Market at a Glance ProtecTIER with HyperFactor DD880 RockSoft Hash-based PRODUCT STABILITY DXi7500 RockSoft Hash-based S2100-ES2 VTL 700 SIR Hash-based DeltaStor Acquired by ! Over $400 ! Small struggling Ø Acquisition or EMC million in debt company failure imminent ProtecTIER in In production production since 2006 since 2006 ! Post process ! GA October 2008 ! GA May 2008 IBM in business for nearly 100 years Over 25PBs of in production CAPACITY-SCALABILTY Single system can scale to 1PB capacity Up to 16 virtual tape libraries Up to 512 virtual tape drives Up to 512,000 virtual tape cartridges Many small systems in production customers customers ! 58TB Maximum !Limited by rapid !Limited by rapid useable capacity hash table growth !Limits not published !Limits not published !Limits not published MEETS ENTERPRISE REQUIREMENTS? YES ! Very few small ! Very few small ! NO hash table growth Ø Almost no deduplication in production See Note (12-13) See Note (14-15) ! Limited by huge storage requirements Up to 64 virtual Up to 128 virtual Up to 192 virtual tape libraries Up to160 virtual tape drives Up to130,000 virtual cartridges ! NO tape libraries tape libraries Up to 1024 Up to 192 virtual virtual drives Up to 64,000 virtual cartridges tape drives Up to 5.3 million virtual cartridges ! NO ! NO 29 © 2011 IBM Corporation