3 Why a tutorial on Scientific Data Management at Pittcon? As scientific research becomes more oriented towards high-volume lab work, there will be increasing problems in managing large volumes of data. Regulatory changes are having an important impact on laboratory data management. It is becoming increasingly important to assure long- term preservation of data of all sorts; techniques developed and understood in the scientific data management area can help. This class was given successfully at previous LIMS conferences
4 The key matter to be discussed today Once you have collected your data, and the data has been written into an output file: –On what storage medium/system, –and in what logical structure, –and with what sort of record of custody should data be stored in to assure its long term readability and utility?
5 Approach & Goals Approach –This tutorial casts a very wide net. –We will cover a large span of scale – ranging from single spreadsheets to hundreds of TBs of data. Goals –Explain the key problems, and the concepts and nomenclature surrounding the problems –Identify some of what might be the right answers, and a few of the answers that are definitely wrong –Provide information and references so you can independently pursue matters of interest to you. –At the end of the tutorial, you might not be ready to design a large data management system, but you will know how to start
6 Sources & format There exists no text for this material that covers this material in the manner discussed in this tutorial. CAS is an expert in some of the areas to be discussed today, but not all. Expect extensive footnoting and acknowledgement of other sources. The level of detail is intentionally uneven. Greater detail is generally associated with one of two factors: –A topic is sufficiently straightforward that some details will let the participant go off and do something on her/his own. –A topic is especially important and is included in the notes in detail so participants may refer to it later. (In this case we may skim over some details during the actual presentation).
7 Outline The problem – the data deluge, plus data doesn't age as gracefully as you (probably) think Physical storage of data: RAID, tapes, CDs, etc. Data security & Legal issues Data management strategies: –Flat files –Excel as a scientific data management tool –Relational databases Specialized scientific data storage formats –Data exchange among heterogeneous formats. –Data warehouses, federations, and grids –Visualization and collection-time data reduction Closing thoughts
8 Bits, Bytes, and the proof that CDs have consciousness A bit is the basic unit of storage, and is always either a 1 or a 0. 8 bits make a byte, the smallest usual unit of storage in a computer. MegaByte (MB) - 1,048,576 bytes (A CD-ROM holds ~ 600 MBs) GigaByte (GB) – ~ 1 billion bytes TeraByte (TB) - ~ 1 trillion bytes (a large library might have ~1 TB of data in printed material) PetaByte (PB) – 1 thousand TBs ExaByte (EB) – 1 thousand PBs
9 The problem of scientific data management
10 Explosion of data and need to retain it Science historically has struggled to acquire data; computing was largely used to simulate systems without much underlying data Lots of data: –Lots of data available “out there” –Dramatically accelerating ability to produce new data One of the key challenges, and one of the key uses of computing, is now to make sense out of data now so easily produced Need to preserve availability of data for ???
11 Accelerating ability to produce new data Diffractometer – 1 TB/year Synchotron – 60 GB/day bursts Gene expression chip readers – 360 GB/day Human Genome – 3 GB/person High-energy physics – 1 PB per year
12 Some things to think about 25 years ago data was stored on punched tape or punched cards How would you get data off an old AppleII+ diskette? How about one of those high-density 5 ¼” DOS diskettes? The backup tape in the sock drawer (especially if it’s a VMS backup tape of an SPSS-VMS data file) The no-longer-easily-handled data file on a CD (e.g Census data) Data is essentially irreproducible more than a short period of time after the fact
13 Have you even tried to read one of your old data files? Exp_2_2_feb_14_
14 Even a small file can be undecipherable! 1 m F F M M F M
15 And something even older… Hwæt! We Gardena in geardagum, þeodcyninga, þrym gefrunon, hu ða æþelingas ellen fremedon. Oft Scyld Scefing sceaþena þreatum… This is from Beowulf, written 1,000 years ago. Think about the language problem relative to the half-life of radioactive waste!
16 Physical storage of data: tapes, CDs, disk
17 Durability of media Stone: 40,000 years Ceramics: 8,000 years Papyrus: 5,000 years Parchment: 3,000 years Paper: 2,000 years Magnetic tape: 10 years (under ideal conditions; 3-5 more conservative) CD-RW: 5-10 years (under ideal conditions; 1.5 years more conservative) Magnetic disk: 5 years Even if the media survives, will the technology to read it?
18 Data storage: media issues So what do you do with data on a paper tape? Long term data storage inevitably forces you to confront two issues: –the lifespan of the media –the lifespan of the reading device Removable Magnetic media –The right answer to any long-term (or even intermediate- term) data storage problem is almost never diskettes. It’s always a race between the lifespan of the media and the lifespan of the readers. –Esoteric removable magnetic media are never a good idea. Even Zip drives are probably not a good bet in the long run. What do you do with a critical data set when your only copy is on a Bernoulli drive?
19 Magnetic Tapes Tapes store data in tracks on a magnetic medium. The actual material on the tape can become brittle and/or worn and fall off. Tapes are best used in machine room environments with controlled humidity. There are three situations in which tapes are the right choice: –Within production machine rooms –As backup media –For transfer between machine rooms under some circumstances
20 Tape formats There are several formats with small user bases; these should probably be avoided. DAT tapes don’t last well For system backups of office, lab, or departmental servers, Digital Linear Tape (DLT) is best choice In machine rooms, Linear Tape Open (LTO) is the best choice. ( LTO is a multi-vendor standard with two variants: –Accelis: faster, lower capacity, lower popularity –Ultrium: Gbps, high capacity (100 GB/tape; 200 w compression). Excellent for write-intensive applications
22 Tape Robots STK Tape Silo –Holds thousands of tapes –2.4 PB total capacity Xcerta tape reader –Holds 10 tapes –600 GB total capacity
23 Tape conversion services If you are presented with data in a physical format you can ’ t read, there are several services that outsource data recovery Do be careful of two issues that may be separate: getting files off a tape for which you have no reader; getting the files into a format you can read with software you have. There are several of these companies. Two examples: –Mueller Media Corporation (capabilities include recovery in a fashion suitable for litigation purposes) –Legacy Engineering
24 Non-magnetic removable media CD – Compact Disk MB CD-ROM – CD-Read Only Memory CD-RW – CD –Read/Write CD speeds: 12x2x24 (x = 150 KB) DVD-R Digital Versatile Disk 4.7 GB DVD-RAM DVD-RW – DVD-Read/Write (Pioneer) DVD+RW – Another DVD Read/Write (Sony/Philips) – the likely future most common standard Don’t set any of these on your dashboard! CD-RW diagram rw.htm#CD-R
25 CDs and DVDs con’t For routine, reliable, reasonably dense storage of data around the lab, you can’t beat CDs or DVDs. CD writers are commonplace & reliable DVD writers are newer, more costly, and more prone to format issues. Always be sure to have extensive and complete information on the CD – including everything you need to know to remember what it really is later. There should be no data physically on the CD that is not contained in a file burned on the CD. Watch out for longevity issues!! –CD R/W – can be rewritten up to 1,000 times –Shelf life 5-10 years
26 CD & DVD Jukeboxes Jukeboxes are good for what they do Because the basic media are standard, if you had to ditch your investment in the jukebox itself you could still reuse the media 240 CD jukebox at left from x.com/index.htm
27 CD & DVD Jukeboxes, con’t System shown at left holds 16 jukeboxes; each holds 240 CDs x.com/index.htm
28 Spinning disk storage JBOD (Just a Bunch of Disk) – alright so long as it’s alright to loose data now and again. High speed access, takes advantage of relatively low cost of disk drives. Good for temporary data parking while data awaits reduction. RAID (Redundant Array of Independent Disks) – what you need if you don’t want to lose data. Lifecycle replacement an issue in both cases
29 Types of disk SCSI (Small Computer System Interface. ATA (Advanced Technology Advancement) or IDE –Intended for “internal to server” use – 40 cm cables –Most people mean ATA when they say IDE (Intelligent Drive Electronics ). Most people also mean Parallel ATA when they say ATA Enhanced IDE, a newer version of IDE developed by Western Digital Corporation, also called ATA-2 Serial ATA –Evolutionary replacement for ATA –Thinner, longer cables – 1 meter Fibre channel – ANSI standard for a machine room fabric connecting disks
30 Disk Trends Capacity: doubles each year Transfer rate: 40% per year MB per $: doubles each year Currently – around $5,000 per GB for cheapest options
31 RAID* Level 0: Provides data striping (spreading out blocks of each file across multiple disks) but no redundancy. This improves performance but does not deliver fault tolerance. Level 1: Provides disk mirroring. Level 3: Same as Level 0, but also reserves one dedicated disk for error correction data. It provides good performance and some level of fault tolerance. Level 5: Provides data striping at the byte level and also stripe error correction information. This results in excellent performance and good fault tolerance.
32 RAID 3 “This scheme consists of an array of HDDs for data and one unit for parity. … The scheme generates from XOR (exclusive- or) parity derived from bit 0 through bit7. If any of the HDDs fail, it restores the original data by an XOR between the redundant bits on other HDDs and the parity HDD. With RAID 3, all HDDs operate constantly. “ stuff.com/ADTX/adtxwhatisraid.html
33 RAID 5 “RAID5 implements striping and parity. In RAID5, the parity is dispersed and stored in all HDDs. …. RAID5 is most commonly used in the products on market these days.” *http://www.studio-stuff.com/ADTX/adtxwhatisraid.html
34 But it takes more than RAID… If you have RAID disk arrays, that provides reliable access to data (within a machine room) so long as you don ’ t loose a disk controller To ensure that your data stays available (so long as you have power), each disk array must be attached to two servers simultaneously
35 NAS and SAN Storage Area Network (SAN) is a high-speed subnetwork of shared storage devices. A storage device is a machine that contains nothing but a disk or disks for storing data. A SAN's architecture works in a way that makes all storage devices available to all servers on a LAN or WAN. A network-attached storage (NAS) device is a server that is dedicated to file sharing through some protocol such as NFS. NAS does not provide any of the activities that a server in a server-centric system typically provides, such as , authentication or file management. … Definitions modified from EMC now offers the best of both worlds!
36 Storage Bricks Group of hard disks inside a sealed box Includes spare disks Typically RAID 5 When one disk fails, one of the spares is put to use When you’re out of spares… Sun seems to have originated this idea
37 Data Security
38 Backups A properly administered backup system and schedule is a must. How often should you back up? More frequently than the amount of elapsed time it takes you to acquire an amount of data that you can’t afford to loose. Backup schedules – full and incremental –Example backup scheduled 1 st Sunday of month: full backup Incremental backups from Sunday on Monday, from Monday on Tuesday, from Tuesday on Wednesday, from Sunday on Thursday, from Thursday on Friday, from Friday on Saturday Incremental from first Sunday on Second Sunday Full backups, stored offsite, every six months RAID disk enhances reliability of storage, but it’s not a substitute for backups
39 Backup Office automated backup systems provide backup against system crashes & viruses. Cost - $100 to $500 or more depending upon capacity Portable backup for Laptops – 5 GB hardcards ~$350 Some backup systems –Omnibak (www.hp.com) –Legato (www.legato.com) –Tivoli (IBM) –For single PCs backup to CD-RW is a real option now!
40 Disaster recovery If your data is too important to lose, then it’s too important to have in just one copy, or have all of the copies in just one location. Natural disasters, human factors (e.g. fire), theft (a significant portion of laptop thefts have data theft as their purpose) can all lead to the loss of one copy of your data. If it’s your only copy…… or the only location where copies are kept… Offsite data storage is essential –Vaulting services –Remote locations of your business –Online backup services are now a real option!
41 Data Security Some percentage of laptop thefts are intentional and aimed at stealing data! Windows XP Professional –Encrypting File System (EFS). But if your account is destroyed or you forget the password... –Recovery Agent provides a secondary account with the ability to recover the data Other systems provide similar features And as before…. The 5 GB hardcard can be a real help
42 Legal ramifications HIPAA (Health Insurance Portability and Accountability Act) –Basically requires that any personally identifiable health data be kept totally secure –Good source of information: FDA 21 CFR Part 11 –Basically requires that any data used in drug development have a full audit trail –Good source of information:
43 Getting rid of data (with certainty!) Deleting files is not enough! Wiping Utilities –Symantec Ghost's gdisk utility (used in combination with the "/diskwipe /dod" flags) (http://enterprisesecurity.symantec.com/products/pr oducts.cfm?productID=3) –Declasfy (http://www.dmares.com/maresware/df.htm#DECLA SFY) Hard disk destruction services –E.g. Webroot ecosafe disk destruction (http://www.webroot.com/wb/products/ecosafe/ind ex.php)
44 Data Management Strategies
45 Data management strategies Flat files Spreadsheets and Statistical software Relational Databases XML Specialized scientific data formats
46 Flat files Nothing beats an ASCII flat file for simplicity ASCII files are not typically used for data storage by commercial software because proprietary formats can be accessed more quickly If you want a reliable way to store data that you will be able to retrieve later reliably (media issues notwithstanding), an ASCII flat file is a good choice.
47 Data Management Strategies: Flat files, II IF you use an ASCII flat file for simple long-term storage, be sure that: –The file name is self-explanatory –There is no information embedded in the file name that is not also embedded in the file –Each individual data file includes a complete data dictionary, explanation of the instrument model and experimental conditions, and explanation of the fields –Lay the data out in accordance with First, Second, and Third Normal Forms as much as is possible (more on these terms later)
48 Data dictionary Definition from webopedia.com: –In database management systems, a file that defines the basic organization of a database. A data dictionary contains a list of all files in the database, the number of records in each file, and the names and types of each field. … More generally: –A data dictionary is what you (or someone else) will need to make sense of the data more than a few days after the experiment is run
49 Spreadsheet Software as a data management tool Microsoft’s Excel may suffice for many data management needs (it is NOT FDA CFR Part 11 compliant!) If any given data set can be described in a 2D spreadsheet with up to hundreds of rows and columns, and if there is relatively little need to work across data sets, then Excel might do the trick for you
50 Spreadsheet software as a data management tool, con’t Designed originally to be electronic accountant ledgers Feature creep in some ways has helped those who have moderate amounts of data to manage There are several options, including Open Source products such as Gnumeric and nearly open source products such as StarOffice (see Since MS Excel is the most commonly used spreadsheet package, this discussion will focus on MS Excel
51 The MS Excel Data menu Sort: Ascending or descending sorts on multiple columns Lists: Allow you to specify a list (use only one list per spreadsheet) and then perform filters, selecting only those that meet a certain criteria (probably more useful for mailing lists than scientific data management) Validation: lets you check for typos, data translation errors, etc. by searching for out of bounds data Consolidate Group and outline Pivottable Get external data
52 MS Excel Statistics Mean, standard deviation, confidence intervals, etc. up to t-test are available as standard functions within MS Excel One-way ANOVA and more complex statistical routines are available in the Statistics Add-in Pack
53 MS Excel Graphics Does certain things quite easily If it doesn’t do what you want it to do easily – it probably won’t do it at all Constraints on the way data are laid out in the spreadsheet are often an issue
54 Statistical Software as a data management tool SPSS and SAS are the two leading packages Both have ‘spreadsheet-like’ data entry or editing interfaces Both have been around a long time, and are likely to remain around for a good while Workstation and mainframe versions of both available
55 What’s wrong with this program? DATA LIST FILE=sample.dat /id 1 v1 3 (A) v2 5 v3 7-9 v4 11 v LIST VARIABLES v1 v2 v3 ONEWAY v3 BY v2 (1,3) REGRESSION /DEPENDENT=v5 /METHOD=ENTER v3 FINISH m f f m m f m
56 Better…. DATA LIST FILE=sample.dat /id 1 gender 3 (A) weight 5 glucose 7-9 bp 11 reactime LIST VARIABLES gender weight glucose ONEWAY glucose BY weight (1,3) REGRESSION /DEPENDENT=reactime /METHOD=ENTER glucose FINISH m f f m m f m
57 Now you have a fighting chance DATA LIST FILE=sample.dat /id 1 gender 3 (A) weight 5 glucose 7-9 bp 11 reactime VARIABLE LABELS ID ‘Subjet ID #' GENDER 'Subject Gender' WEIGHT ‘Subject Weight in pounds’ GLUCOSE ‘Blood glucose level’ BP ‘Blood Pressure’ REACTIME ‘Reaction Time in Minutes” VALUE LABELS GENDER m ‘Male’ f ‘Female’ LIST VARIABLES gender weight glucose ONEWAY glucose BY weight (1,3) REGRESSION /DEPENDENT=reactime /METHOD=ENTER glucose FINISH 1 m f f
58 An example SAS program /* Computer Anxiety in Middle School Chlidren */ /* The following procedure specifies value lables for variables */ PROC FORMAT; VALUE $sex 'M'='Male' 'F'='Female'; VALUE exp 1='upto 1 year' 2='2-3 yrs' 3='3+ yrs'; VALUE school 1='rural' 2='city' 3='suburban'; DATA anxiety; INFILE clas; INPUT ID 1-2 SEX $ 3 (EXP SCHOOL) (1.) (C1-C10) (1.) (M1-M10) (1.) MATHSCOR COMPSCOR 28-29; FORMAT SEX $SEX.; FORMAT EXP EXP.; FORMAT SCHOOL SCHOOL.; /* conditional transformation */ IF MATHSCOR=99 THEN MATHSCOR=.; IF COMPSCOR=99 THEN COMPSCOR=.; /* Recoding variables. Several items are to be reversed while scoring. */ /* The Likert type questionnaire had a choice range of 1-5 */ C3=6-C3; C5=6-C5; C6=6-C6; C10=6-C10; M3=6-M3; M7=6-M7; M8=6-M8; M9=6-M9; COMPOPI = SUM (OF C1-C10) /*FIND SUM OF 10 ITEMS USING SUM FUNCTION */; MATHATTI = M1+M2+M3+M4+M5+M6+M7+M8+M9+M10 /*ADDING ITEM BY ITEM */; /* Labeling variables */ LABEL ID='STUDENT IDENTIFICATION' SEX='STUDENT GENDER' EXP='YRS OF COMP EXPERIENCE' SCHOOL='SCHOOL REPRESENTING' MATHSCOR='SCORE IN MATHEMATICS' COMPSCOR='SCORE IN COMPUTER SCIENCE' COMPOPI='TOTAL FOR COMP SURVEY' MATHATTI='TOTAL FOR MATH ATTI SCALE';
59 SAS example, Part 2 /* Printing data set by choosing specific variables */ PROC PRINT; VAR ID EXP SCHOOL MATHSCOR COMPSCOR COMPOPI MATHATTI; TITLE 'LISTING OF THE VARIABLES'; /* Creating frequency tables */ PROC FREQ DATA=ANXIETY; TABLES SEX EXP SCHOOL; TABLES (EXP SCHOOL)*SEX; TITLE 'FREQUENCY COUNT'; /* Getting means */ PROC MEANS DATA=ANXIETY; VAR COMPOPI MATHATTI MATHSCOR COMPSCOR; TITLE 'DESCRIPTIVE STATICTS FOR CONTINUOUS VARIABLES'; RUN; /* Please refer to the following URL for further infomation */ /* */
60 An example SPSS program TITLE 'COMPUTER ANXIETY IN MIDDLE SCHOOL CHILDREN' DATA LIST FILE=clas.dat /ID 1-2 SEX 3 (A) EXP 4 SCHOOL 5 C1 TO C M1 TO M MATHSCOR COMPSCOR MISSING VALUES MATHSCOR COMPSCOR (99) RECODE C3 C5 C6 C10 M3 M7 M8 M9 (1=5) (2=4) (3=3) (4=2) (5=1) RECODE SEX ('M'=1) ('F'=2) INTO NSEX /* Changing char var into numeric var COMPUTE COMPOPI=SUM (C1 TO C10) /*Find sum of 10 items using SUM function COMPUTE MATHATTI=M1+M2+M3+M4+M5+M6+M7+M8+M9+M10 /* Adding eachi item VARIABLE LABELS ID 'STUDENT IDENTIFICATION' SEX 'STUDENT GENDER' EXP 'YRS OF COMP EXPERIENCE' SCHOOL 'SCHOOL REPRESENTING' MATHSCOR 'SCORE IN MATHEMATICS' COMPSCOR 'SCORE IN COMPUTER SCIENCE' COMPOPI 'TOTAL FOR COMP SURVEY' MATHATTI 'TOTAL FOR MATH ATTI SCALE'
61 SPSS Example, Part 2 /*Adding labels VALUE LABELS SEX 'M' 'MALE' 'F' 'FEMALE'/ EXP 1 'UPTO 1 YR' 2 '2 YEARS' 3 '3 OR MORE'/ SCHOOL 1 'RURAL' 2 'CITY' 3 'SUBURBAN'/ C1 TO C10 1 'STROGNLY DISAGREE' 2 'DISAGREE' 3 'UNDECIDED' 4 'AGREE' 5 'STRONGLY AGREE'/ M1 TO M10 1 'STROGNLY DISAGREE' 2 'DISAGREE' 3 'UNDECIDED' 4 'AGREE' 5 'STRONGLY AGREE'/ NSEX 1 'MALE' 2 'FEMALE'/ PRINT FORMATS COMPOPI MATHATTI (F2.0) /*Specifying the print format comment Listing variables. * listing variables. LIST VARIABLES=SEX EXP SCHOOL MATHSCOR COMPSCOR COMPOPI MATHATTI/ FORMAT=NUMBERED /CASES=10 /* Only the first 10 cases FREQUENCIES VARIABLES=SEX,EXP,SCHOOL/ /* Creating frequency tables STATISTICS=ALL USE ALL. ANOVA COMPSCOR by EXP(1,3). FINISH comment Please refer to the following URL for further infomation
62 Keys to using Statistical Software as a data management tool Be sure to make your programs and files self-defining. Use variable labels and data labels exhaustively. Write out ASCI versions of your program files and data sets. Stat packages generally are able to produce platform- independent ‘transport’ files. Good for transport, but be wary of them as a long-term archival format Statistical software is excellent when your data can be described well without having to use relational database techniques. If you can describe the data items as a very long vector of numbers, you’re set! Statistical software is especially useful when many transformations or calculations are required - but beware transforms, calculations, and creation of new variables interactively!
63 Your own applications in Perl or C Perl –Portable extensible report language –Problematic esoteric rubbish lister –It’s a bit of both –Perl is good way to manipulate small amounts of data in a prototype setting, but performance in a production setting will probably seem inadequate Use Perl to prototype, but if you’re using Perl, rewrite the final application in C or C++
64 LIMS systems The opposite of data reduction…. Developed for petrochemical and pharmaceutical applications –Highly repetitive tests –Regular comparisons with standards –Legal compliance issues are often involved If you need a LIMS system, good rule of thumb is 10X expansion of storage needs Assume a LIMS system will require at least 0.5 FTE dedicated staff for a lab or lab group
65 LIMS systems, con ’ t Sapphire (Made by LabVantage). –One of the standard large LIMS –Very good on regulatory compliance Nautilus (Made by Thermo Electron Corp 5,10380,00.html) –Good LIMS system, perhaps the best of the easier LIMS to use Good source of review information: LIMSource
66 Laboratory Electronic Notebook Intuitively similar function – computerizing lab processes The concept is that a LEN should be less constraining than a LIMS Results thus far are mixed Two example systems –Tripos Electronic Notebook Tech/enterpriseInfo/opInfo Tech/ten.html –DOE 2000 Electronic Notebook note/
67 Database Definitions Database management system: A collection of programs that enables you to store, modify, and extract information from a database. Types of DBMSs: relational, network, flat, and hierarchical. If you need a DBMS, you need a relational DBMS Query: a request to extract data from a database, e.g.: –SELECT ALL WHERE NAME = “JONES" AND AGE > 21 SQL (structured query language) – the standard query language
68 Relational Databases* Relational Database theory developed at IBM by E.F. Codd (1969) Codd's Twelve Rules – the key to relational databases but also good guides to data management generally. Codd’s work is available in several venues, most extensively as a book. The number of rules has now expanded to over 300, but we will start with rules 1-12 and the 0th rule. 0th rule: A relational database management system (DBMS) must manage its stored data using only its relational capabilities. *Based on Tore Bostrup.
69 Codd’s 12 rules 1. Information Rule. All information in the database should be represented in one and only one way -- as values in a table. 2. Guaranteed Access Rule. Each and every datum (atomic value) is guaranteed to be logically accessible by resorting to a combination of table name, primary key value, and column name. 3. Systematic Treatment of Null Values. Null values (distinct from empty character string or a string of blank characters and distinct from zero or any other number) are supported in the fully relational DBMS for representing missing information in a systematic way, independent of data type.
70 Codd’s 12 rules, con’t 4. Dynamic Online Catalog Based on the Relational Model. The database description is represented at the logical level in the same way as ordinary data, so authorized users can apply the same relational language to its interrogation as they apply to regular data.
71 Codd’s 12 rules, con’t 5. Comprehensive Data Sublanguage Rule. A relational system may support several languages and various modes of terminal use. However, there must be at least one language whose statements are expressible, per some well-defined syntax, as character strings and whose ability to support all of the following is comprehensible: –data definition –view definition –data manipulation (interactive and by program) –integrity constraints –authorization –transaction boundaries (begin, commit, and rollback).
72 Codd’s 12 rules, con’t 6. View Updating Rule. All views that are theoretically updateable are also updateable by the system. 7. High-Level Insert, Update, and Delete. The capability of handling a base relation or a derived relation as a single operand applies not only to the retrieval of data, but also to the insertion, update, and deletion of data. 8. Physical Data Independence. Application programs and terminal activities remain logically unimpaired whenever any changes are made in either storage representation or access methods.
73 Codd’s 12 rules, con’t 9. Logical Data Independence. Application programs and terminal activities remain logically unimpaired when information preserving changes of any kind that theoretically permit unimpairment are made to the base tables. 10. Integrity Independence. Integrity constraints specific to a particular relational database must be definable in the relational data sublanguage and storable in the catalog, not in the application programs.
74 Codd’s 12 rules, con’t 11. Distribution Independence. The data manipulation sublanguage of a relational DBMS must enable application programs and terminal activities to remain logically unimpaired whether and whenever data are physically centralized or distributed. 12. Nonsubversion Rule. If a relational system has or supports a low-level (single-record-at-a-time) language, that low-level language cannot be used to subvert or bypass the integrity rules or constraints expressed in the higher-level (multiple-records-at-a- time) relational language.
75 The problem with (some) DBMS computer science Database theory is wonderful stuff It is sometimes possible to get so caught up in the theory of how you would do something that the practical matters of actually doing it go by the wayside This is particularly true of the concept of “normal forms” – only three of which we will cover
76 Some terminology A key is a field that *could* serve as a unique identifier of records. The Primary key is the one field chosen to be the unique identifier of records.
77 First Normal Form Reduce entities to first normal form (1NF) by removing repeating or multivalued attributes to another, child entity. Specimen #Measurement#Value Specimens 14
78 Second Normal Form Reduce first normal form entities to second normal form (2NF) by removing attributes that are not dependent on the whole primary key.
79 Third Normal form Reduce second normal form entities to third normal form (3NF) by removing attributes that depend on other, nonkey attributes (other than alternative keys). It may at times be beneficial to stop at 2NF for performance reasons!
80 On to database products Microsoft Access – Common, relatively inexpensive, moderately scalable. O.k. for personal use Microsoft SQL Server – More scalable – commonly used for departmental (or larger) databases Oracle – Common, relatively more expensive, extremely robust and scalable DB2 – Relatively common, IBM’s commercial database application MySQL – Becoming more common, free, good for prototyping and small-scale applications
81 MySQL Open source database software Available for several operating systems Downloadable from Excellent for prototyping database applications, and in many cases plenty for production
82 Components of MySQL (exemplary of database products generally) mysql – executes sql commands mysqlaccess – manages users mysqladmin – database administration mysqld – MySQL server process mysqldump – dumps definition and contents of a database into a file mysqlhotcopy – hot backup of databast mysqlimport – imports data from other formats mysqlshow – shows information about server and objects mysqld_safe – starts and manages mysql on Unix
83 Database applications and the web? An Open Source option –MySQL - database –PHP - web scripting application –Apache - web server Oracle and its web modules Stat package and web modules
84 XML The Extensible Markup Language (XML) is the universal format for structured documents and data on the Web. Half of “XML in 10 points” (http://www.w3.org/XML/1999/XML-in-10-points) –XML is for structuring data. XML makes it easy for a computer to generate data, read data, and ensure that the data structure is unambiguous. –XML looks a bit like HTML. Like HTML, XML makes use of tags (words bracketed by ' ') and attributes (of the form name="value"). –XML is text, but isn't meant to be read. –XML is verbose by design. (And it’s *really* verbose) –XML is a family of technologies. (This leads to the opportunity to create discipline-specific XML templates)
85 XML XML really is one of the most important data presentation technologies to be developed in recent years XML is a meta-markup language The development and use of DTDs (document type definition) is time consuming, critical, and subject to the usual laws regarding standards XML is a way to present data, but not a good way to organize lots of data
86 Some XML examples Chemical Markup Language Extensible Data Format CellML Chemical Markup Language SBML (Systems Biology Markup Language) Extensible Data Format MathML –(a + b) 2 (from re.htm) a + b 2
87 XML issues Great technology Good commercial authoring systems available or in development The problem with standards…. Perhaps the biggest challenge in XML is the fact that it is so easy to put together a web site and propose a DTD as a standard, making the creation of real standards a challenge
88 XML vs PDF PDF files are essentially universally readable. PDF file formats give you a picture of what was once data in a fashion that makes retrieval of the data hard at best. XML requires a bit more in terms of software, but preserves the data as data, that others can interact with. Utility of XML and PDF interacts with proprietary concerns, institutional concerns, and community concerns – which are not always in harmony!
89 Specialized data storage formats - HDF Hierarchical Data Format (HDF) HDF is an open-source effort HDF5 is a general purpose library and file format for storing scientific data.
90 HDF, con’t HDF5 can store two primary objects: datasets and groups. A dataset is essentially a multidimensional array of data elements, and a group is a structure for organizing objects in an HDF5 file. Using these two basic objects, one can create and store almost any kind of scientific data structure. Designed to address the data management needs of scientists and engineers working in high performance, data intensive computing environments. HDF5 emphasizes storage and I/O efficiency. HDF is nontrivial to implement If you need the full capabilities of HDF, there’s nothing like it
91 Free Software Foundation Many of the software products mentioned in this talk (XML, Perl, etc.) are Open Source Software The GNU general public license is the standard license for such software Some of the best software for specific scientific communities is open source (community software) There are certain expectations about such software and how it is used
92 Data exchange among heterogeneous formats I have data files in SAS, SPSS, Excel, and Access formats. What do I do? Each of the more widely used stat packages contain significant utilities for exchanging data. Stata makes a package called Stat Transfer DBMS/Copy (Conceptual Software) probably the best software for exchange among heterogeneous formats
93 Distributed Data Data warehouses Data federations Distributed File Systems External data sources Data Grids
94 Data warehouses In a large organization one might want to ask research questions of transactional data. And what will the MIS folks say about this? Transactions have to happen now; the analysis does not necessarily have to. Data warehousing is the coordinated, architected, and periodic copying of data from various sources, both inside and outside the enterprise, into an environment optimized for analytic and informational processing (Definition from “Data warehousing for dummies” by Alan R. Simon
95 Getting something out of the data warehouse Querying and reporting: tell me what’s what OLAP (On-Line Analytical Processing): do some analysis and tell me what’s up, and maybe test some hypotheses Data mining: Atheoretic. Give me some obscure information about the underlying structure of the data EIS (Executive Information Systems): boil it down real simple for me
96 More Buzzwords Data Mart: Like a data warehouse, but perhaps more focused. [Term often used by the newly renamed Data Mart team after a Data Warehouse fiasco] Operational Data Store: Like a data warehouse, but the data are always current (or almost). [Day traders]
97 Distributed File Systems - OpenAFS The file system formerly known as Andrew File System – Widely used among physicists AFS is a distributed filesystem product, pioneered at Carnegie Mellon University and supported and developed as a product by Transarc Corporation (now IBM Pittsburgh Labs). It offers a client-server architecture for file sharing, providing location independence, scalability and transparent migration capabilities for data. The only show in town for simple data distribution without going to experimental computer science projects or fairly involved commercial products
98 AFS Structure AFS operates on the basis of “cells” Each cell depends upon a cell server that creates the root level directory for that cell Other network-attached devices can attach themselves into the AFS cell directory structure Moving data from one place to another than becomes just like a file operation except that it is mediated by the network Requires installation of client software (available for most Unix flavors and Windows) Root server Client 1Department 1Department 2 Researcher 1Researcher 2
99 Grids What’s a grid? Hottest current buzzword A way to link together disparate, geographically disparate computing resources to create a meta-computing facility The term ‘computing grid’ was coined in analogy to the electrical power grid Three types of grids: –Compute –Collaborative –Data
100 Compute Grids Compute grids tie together disparate computing facilities to create a metacomputer. Supercomputers: Globus is an experimental system that historically focuses on tying together supercomputers PCs: –Entropia is a commercial product that aims to tie together multiple PCs
101 Collaboration Grids
102 Data Grids - Example Tier 0: CERN Tier 1: A national center Tier 2 a center covering one region of a large country Tier 3: workgroup server Tier 4: the (thousands of) desktops
103 Example Data Grids GriPhyN (Grid Physics Network) – The key problem: too much data (PB per year) Biomedical data Globus – beginning to integrate data grid functionality Avaki – commercial data grid product Data Grids “virtualize” data locality Chemistry example – Reciprocal Net (http://www.reciprocalnet.org/)
106 Web-accessible databases Especially prominent in biomedical sciences. E.g. NCBI: Entrez Pubmed –http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=PubMed –Provides access to over 11 million MEDLINE citations Nucleotide –http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=Nucleotide –collection of sequences from several sources, including GenBank, RefSeq, and PDB. Protein –http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=Protein Genome –http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=Genome –The whole genomes of over 800 organisms.
107 Federated databases A federation of databases is a group of databases that are tied together in some reasonable way permitting data retrieval (generally) and sometimes (maybe in the future) data writing Benefits of federated approach: –Local access control. Lets data owner control access –Acknowledges multiple sources of data –By focusing on the edges of contact, should be more flexible over the long run Shortcomings: Right now, significant hand work in constructing such systems Example product: IBM’s DiscoveryLink
108 The idealized view of DiscoveryLink Architecture Clinical Data Toxicity Data Lab Results DL
110 Microarray Data Portal Web application and database designed for annotation and analysis of microarray experiments. Annotation: Designed for users to set up experimental design first minimizing amount of time for sample entry but still getting in the essential info Analysis –Allows user to partition data into groups based on their annotation. –Extensive filtering, search, and display options –T-test, Clustering, SVD, etc. –Allows different views of data based on informatics associated with the genes (e.g. KEGG, GO, Chromosome Location)
112 KEGG pathway information
113 Online Biological Data Retrieval Web queries used to quickly identify SNPs and Genes in specific regions and return information about those identified SNPs and Genes. Used by the Hereditary Diseases and Family Studies Division of the Medical and Molecular Genetics Department of the Indiana University School of Medicine. Live demo (hopefully) Marker1: D5S2057 Marker2: D5S436 Filter on tissue expression: Muscle < 60 seconds vs 10 hours
115 A commercial data grid: Avaki Provides a set of features that are similar to other data grid projects described Provides excellent security Economies of scale Becoming widely used in life sciences
116 Real-time data reduction as a critical strategy Data: bits and bytes Information: that which reduces uncertainty (Claude Shannon). Literally …the difference between two forms of organization or between two states of uncertainty before and after a message has been received, but also the degree to which one variable of a system depends on or is constrained by another. (from In other words, if there is no realistic circumstance in which you would take an action based on or influenced by a certain number, than this number is data, not information We collect a lot more data than we do information!
117 Real-time data reduction Given that we collect much more data than information, what do we do? If we can identify something as reliably just data, and definitely not possibly information, why keep it? In some cases of instruments that produce data continually, a PC dedicated to on-the-fly data reduction can drastically reduce data storage requirements
118 Knowledge management, searchers, and controlled vocabularies A tremendous amount of effort has gone in to natural language processing, AI, knowledge discovery, etc. with results ranging from mixed to disappointing. If you want to be able to search large volumes of data on an ad-hoc basis, then controlled vocabularies are essential. Results here are mixed as well, but at least the problems are sociological, not technological. Examples: –GO (Gene Ontology) Gene Ontology Consortium, –MeSH (Medical Subject Headings)
119 Visualization The days when you could take a stack of greenbar down to your favorite bar, page through the output, and understand your data are gone. Data visualization is becoming the only means by which we can have any hope of understanding the data we are producing A single gene expression chip can produce more pixels of data than the human eye&mind together are capable of processing
120 Gene expression chips
121 Visualization Options For 2D: your monitor and some software! 2D commercial software 2D Open source: OpenmDX
122 Large-scale 3D systems CAVE™ - Cave Automatic Virtual Environment –Anything *but* automatic –Best immersive 3D technology available Immersadesk –Furniture-scale 3-D environment –Easier to program than CAVE –Immersive 3D feel not as good as CAVE, but less expensive
A Lab-scale 3D system – the John- E-Box TM Commercially available from CAE-Net, Inc.
124 Heirarchical Storage Management Systems Differential cost of media –RAM$60-$100/MB –RAID$4-$10/MB –CD~$1/MB (readers included) –Tape$0.05-$1/MB Differential read rates and access times: –Disk: 1 GB/sec; 9-20 ms access time –Tape: 200 MB/sec; <1 min (autoloader)
125 Hierarchical Storage Management The objective of an HSM is to optimize the distribution of data between disk and tape so as to store extremely large amounts of data at reasonably economical costs while keeping track of everything Most data is read rarely. Tape is cheap. Keep rarely read data on disk. Data that is often used keep on disk. Stage data to disk on command for faster access when you know you’re going to need it later. Stage data to disk in output. Manage data on tape so as to handle security and reliability. Metadata system keeps track of what everything is and where it is!
126 HSM products EMASS Inc. - AMASS (Archival Management and Storage System). Veritas – LSF – Sun Microsystems, Inc. HPSS (High Performance Storage System) – a consortium-lead product designed originally for weapons labs and now marketed by IBM Tivoli storage Manager ibm.com/software/tivoli/products/storage-mgr/
127 HPSS – High Performance Storage System Controlled by a consortium, but produced and released as a service from IBM (as opposed to a product) Designed to meet the needs of some of the most demanding and security-conscious customers in the world Customers include: –Lawrence Berkely Laboratories –Los Alamos National Laboratories –Sandia National Laboratories –San Diego Supercomputer Center –Indiana University
128 More about HPSS Requirements –Absolute reliability of data in all forms (reliably read whenever authorized person wants, and reliably not available to anyone unauthorized) –High capacity –Speed –Fault detection/Correction Components –Name Server (NS) – translates standard file names and paths into HPSS object identifier –Bitfile Server (BFS) – provides logical bitfiles to clients –Storage Server (SS) – manages relationship between logical files and physical files –Physical Volume Library (PVL) – maps logical volumes to physical cartridges. Issues commands to PVR –Physical Volume Repository – mounts and dismounts cartridges –Mover (MVR) – transfers data from a source to a sink
130 The future of storage “In-place” increases in density New technologies: –WORM Optical Storage & holographics –Millepedes –Non-corrosive metal
132 Millipede Storage Based on atomic force microscopy (AFM): tiny depressions melted by an AFM tip into a polymer medium represent stored data bits that can then be read by the same tip. Thermomechanical storage is capable of achieving data densities in the hundreds of Gb/in² range Current best – 20 to 100 Gb/in² Expected limits for magnetic recording (60–70 Gb/in²).
133 Millipede Storage, Part 2 Read/Write rate of individual probe is limited The Read/Write head consists of ~1,000 individual probes that read in parallel
134 Storage of text on nonreactive metal disks All of the commonly used storage media depend upon arbitrary standards and are fragile If you have data that you really want to keep secure for a long time, why not write it as text on non-corrosive metal disks?
135 Future of computing The PC market will continue to be driven largely by home uses (esp games) In scientific data management, the utility of computing systems will be less determined by chip speeds and more by memory and disk configurations, and internal and external bandwidth And the future is uncertain! –If you can see clearly what your storage requirements are 25 years into the future, and they are large scale and significant, then a tremendous investment based on what’s available today may be reasonable. –In any other case, it may be best to take shorter views – 5 to perhaps 10 years, and build into your thinking the constant need to refresh
136 The ongoing challenge One of the key problems in data storage is that you can’t just store it. Data stored and left alone is unlikely under most circumstances to be readable – and less likely to be comprehensible and useable – in 20 years. The problem, of course, is that there is an ever increasing need for tremendous longevity in the utility of data. Because of this it is essential that data receive ongoing curation, and migration from older media and devices to newer media and devices. Only in this way can data remain useful year after year.
137 A few pointers to references Statistical software: tutorials on Alan R. Simon. Data warehousing for Dummies IDG Books E.R. Harold & W. Scott Means XML in a nutshell. O ’ Reilley A. Khurshudov The essential guide to computer data storage. Prentice Hall A. Barrows Access 2002 for Dummies. IDG G.M. Nielson, H. Hagen, H. Mueller Scientific Visualization. IEEE Computer Society C. Gibas & P. Jambeck Developing bioinformatics computer skills. O ’ Reilly