Geek out 2014-lagergren-final
7 likes•2,959 views
The document discusses the relationship between hardware, operating systems, and Java Virtual Machines over time. It describes how early systems in the 1970s-1990s attempted to execute bytecode natively but never fully implemented it. Intermediate languages have existed since the 1960s to abstract code from hardware. The Java Virtual Machine model introduced in 1995 provided memory protection, type checking, and a "sandbox" environment for Java bytecode, influencing many later virtual machines. Overall, the JVM takes on operating system-like roles through threads, memory management, and locking, while still interfacing with the underlying OS.
Geek out 2014-lagergren-final
- 1. A love/hate rela,onship: The JVM/OS dialec,c Marcus Lagergren, Oracle
- 2. A love/hate rela,onship: The JVM/OS dialec,c Marcus Lagergren, Oracle and the hardware…
- 3. A love/hate rela,onship: The JVM/OS dialec,c Marcus Lagergren, Oracle Fredrik Öhrström, Spo,fy and the hardware…
- 4. The Legal Slide "THE FOLLOWING IS INTENDED TO OUTLINE OUR GENERAL PRODUCT DIRECTION. IT IS INTENDED FOR INFORMATION PURPOSES ONLY, AND MAY NOT BE INCORPORATED INTO ANY CONTRACT. IT IS NOT A COMMITMENT TO DELIVER ANY MATERIAL, CODE, OR FUNCTIONALITY, AND SHOULD NOT BE RELIED UPON IN MAKING PURCHASING DECISION. THE DEVELOPMENT, RELEASE, AND TIMING OF ANY FEATURES OR FUNCTIONALITY DESCRIBED FOR ORACLE'S PRODUCTS REMAINS AT THE SOLE DISCRETION OF ORACLE."
- 5. Agenda • In the borderlands between hardware, OS and JVM, both good and bad things happen • Computer history • How do they affect each other? • Where is it all going?
- 6. Who am I? @lagergren
- 7. Who am I?
- 8. Who am I? mlvm-‐[email protected] nashorn-‐[email protected] hRps://avatar.java.net
- 9. Who am I?
- 10. Who am I?
- 11. Who is Fredrik?
- 12. Who is Fredrik?
- 13. Who is Fredrik?
- 14. Who is Fredrik?
- 16. The Past
- 17. The Past (Skipping very quickly over a tremendous amount of hardware)
- 18. Texas Instruments TI 99/4a • 1979-‐1984 • Contains an OS and a GPL interpreter • Device drivers (DSRs) could be wriRen in GPL
- 19. Texas Instruments TI 99/4a • 1979-‐1984 • Contains an OS and a GPL interpreter • Device drivers (DSRs) could be wriRen in GPL • They intended to execute GPL bytecode na,vely – But they never did
- 20. VIC 20 and Commodore 64 • 1980-‐1986
- 21. VIC 20 and Commodore 64 • 1980-‐1986 http://codebase64.org
- 22. VIC 20 and Commodore 64 • 1980-‐1986
- 23. VIC 20 and Commodore 64 • 1980-‐1986 • Basic interpreter wriRen in C on my Mac – 1000x faster than a physical 6502
- 24. Nothing new under the sun?
- 25. …and stuff MITS Altair 8800, Commodore PET 2001, Apple II, Atari VCS, Tandy Radio Shack TRS-‐80, ABC 80, NASCOM-‐1, Sharp MZ-‐80k, Atari 400/800, MaRel Intellivision, Tangerine Microtan 65, HP-‐85, Sinclair ZX80, Acorn Atom, Sinclair ZX81, Osborne 1, IBM PC, BBC Micro, Sinclair ZX Spectrum, Coleco Vision, GCE/MB Vectrex, Grundy Newbrain, Dragon 32, Jupiter ACE, Compaq Portable, Apple Lisa, Oric 1, MaRel Aquarius, Nintendo Famicom/NES, Acorn Electron, Sony MSX, Apple Macintosh, Sinclair QL, Amstrad CPC-‐464, IBM PC AT, Tatung Einstein, Atari ST, Commodore Amiga, Amstrad PCW, Sega Master System, Acorn Archimedes, NeXT
- 26. The JavaSta,on • 1996-‐2000 • Contains JavaOS, a micro kernel in C with an interpreter • Device drivers were wriRen in Java
- 27. The JavaSta,on • 1996-‐2000 • Contains JavaOS, a micro kernel in C with an interpreter • Device drivers were wriRen in Java • They intended to execute bytecode na,vely – But they never did
- 29. Intermediate Languages • 1966: O-‐code (BCPL) • 1970: p-‐code (Pascal) • 1979: GPL • 1995: Java Bytecode
- 30. Intermediate Languages • 1966: O-‐code (BCPL) • 1970: p-‐code (Pascal) • 1979: GPL • 1995: Java Bytecode
- 31. Intermediate Languages • 1966: O-‐code (BCPL) • 1970: p-‐code (Pascal) • 1979: GPL • 1995: Java Bytecode
- 32. Intermediate Languages • 1966: O-‐code (BCPL) • 1970: p-‐code (Pascal) • 1979: GPL • 1995: Java Bytecode
- 33. …and more stuff Ac,onscript, Adobe Flash objects, BANCStar, CLISP, CMUCL, CLR/.NET, Dalvik, Dis, EiffelStudio, Emacs eLisp-‐>bytecode, Embeddable Common Lisp, Erlang/BEAM, Icon, Unicon, Infocom/Z-‐machine text adventure games, LLVM, Lua, m-‐code/MATLAB, OCaml, Parrot Virtual Machine, R, Scheme 48, Smalltalk, SPIN/Parallax Propeller Microcontroller, SWEET16/Apple II Basic ROM, Visual FoxPro bytecode, YARV, Rubinius
- 34. Intermediate Languages • 1966: O-‐code (BCPL) • 1970: p-‐code (Pascal) • 1979: GPL • 1995: Java Bytecode
- 35. Intermediate Languages • 1966: O-‐code (BCPL) • 1970: p-‐code (Pascal) • 1979: GPL • 1995: Java Bytecode – memory protec,on, type and control verifica,on and explicit security management, “sandbox” model, “object orienta,on”
- 36. Intermediate Languages • 1966: O-‐code (BCPL) • 1970: p-‐code (Pascal) • 1979: GPL • 1995: Java Bytecode – memory protec,on, type and control verifica,on and explicit security management, “sandbox” model, “object orienta,on” • 1999: The JavaOS is discon,nued
- 39. To put it simply • The JVM has OS-‐like behavior – Threads – Memory management/protec,on – Locking • All this is somewhat mi,gated through libc & other libraries
- 40. Threads
- 41. Threads • Heavy weight processes – Slow switching – fork()
- 42. Threads • Heavy weight processes – Slow switching – fork() • Green threads – Fast switching, difficult to implement • Na,ve Locks?
- 43. Threads • Heavy weight processes – Slow switching – fork() • Green threads – Fast switching, difficult to implement • Na,ve Locks? • MxN threads – Even more difficult to implement
- 44. Threads • The old standby: OS Threads • No support for stack overflow • By defini,on, no memory protec,on between threads
- 45. Locks
- 46. Locks
- 48. Thin Locks public class PseudoSpinlock { private static final int LOCK_FREE = 0; private static final int LOCK_TAKEN = 1; public void lock() { //burn cycles while (cmpxchg(LOCK_TAKEN, &lock) == LOCK_TAKEN) { micropause(); //optional } } public void unlock() { int old = cmpxchg(LOCK_FREE, &lock); //guard against recursive locks assert(old == LOCK_TAKEN); } }
- 49. Thin Locks • Use whatever atomic support there is in the hardware / OS • Cheap to lock and unlock, expensive to keep locked
- 50. Fat Locks • Use OS lock support • Expensive to lock and unlock, cheap to keep locked • Need for more advanced synchroniza,on mechanisms – wait – notify
- 51. Adap,ve Behavior • Profile based transmuta,on of thin locks to fat locks – …and vice versa – Nothing your C program can do
- 52. Adap,ve Behavior • Profile based transmuta,on of thin locks to fat locks – …and vice versa – Nothing your C program can do • Biased locking
- 53. Adap,ve Behavior • Profile based transmuta,on of thin locks to fat locks – …and vice versa – Nothing your C program can do • Biased locking • Thread switching heuris,cs / Cache warmup
- 54. Adap,ve Behavior • Constant tension between OS switching and Java switching. – One example of a JVM/OS baRle
- 55. Na,ve Memory
- 56. Na,ve Memory
- 57. Na,ve Memory
- 58. Na,ve Memory Tracking HotSpot: java –XX:NativeMemoryTracking=<summary|detail> Test jcmd <pid> VM.native_memory JRockit: USE_OS_MALLOC=0 TRACE_ALLOC_SITES=1 java Test jrcmd <pid> print_memusage
- 59. Memory Models
- 60. Memory Models public class WhileLoop { //can be accessed by other threads private boolean finished; while (!finished) { do something… } }
- 61. Memory Models public class WhileLoop { //can be accessed by other threads private boolean finished; boolean tmp = finished; while (!tmp) { do something… } }
- 62. Memory Models public class WhileLoop { //can be accessed by other threads private volatile boolean finished; while (!finished) { do something… } }
- 63. Memory Models volatile int x; int y; volatile boolean finished; x = 17; y = 4711; finished = true; if (finished) { System.err.println(x); System.err.println(y); }
- 64. Memory Models volatile int x; int y; volatile boolean finished; x = 17; y = 4711; finished = true; if (finished) { System.err.println(x); System.err.println(y); }
- 65. Memory Models public class GadgetHolder { private Gadget theGadget; public synchronized Gadget getGadget() { if (this.theGadget == null) { this.theGadget = new Gadget(); } return this.theGadget; } }
- 66. Memory Models public class GadgetHolder { private Gadget theGadget; public Gadget getGadget() { if (this.theGadget == null) { synchronized(this) { if (this.theGadget == null) { this.theGadget = new Gadget(); } } } return this.theGadget; } }
- 67. Memory Models public class GadgetHolder { private volatile Gadget theGadget; public Gadget getGadget() { if (this.theGadget == null) { synchronized(this) { if (this.theGadget == null) { this.theGadget = new Gadget(); } } } return this.theGadget; } }
- 68. Memory Models public class GadgetMaker { public static Gadget theGadget = new Gadget(); }
- 69. WTF?
- 70. Taking Control of the OS
- 71. Taking Control of the OS • Taking control of the OS • Taking control of the na,ve memory • Taking control of the C heap
- 72. Taking Control of the OS • Taking control of the OS • Taking control of the na,ve memory • Taking control of the C heap • Well you can’t really, but you can do your best
- 73. Is the JVM an OS? • JRockit Virtual Edi,on • Azul • Cloudius • Jnode • …
- 74. Is the JVM an OS? • JRockit Virtual Edi,on • Implemented libc, libraries and the OS – Not much required for a single process Java OS. • Finally, the Java OS?
- 75. Is the JVM an OS?
- 77. Is the JVM an OS?
- 78. Is the JVM an OS? • Add a coopera,ve aspect to thread switching • Zero-‐copy networking code • Reduce cost of entering OS • Balloon driver • Runs only on hypervisor
- 79. Pauseless GC • Hope that a lot of data is thread local and remains thread local – (it usually is)
- 80. Pauseless GC • Hope that a lot of data is thread local and remains thread local – (it usually is) • Use one large global heap and one thread local heap per thread
- 81. Pauseless GC • Hope that a lot of data is thread local and remains thread local – (it usually is) • Use one large global heap and one thread local heap per thread • If thread local data is exposed to another thread – promote to global heap
- 82. Pauseless GC • We need a write barrier //x.field = y void checkWriteAccess(Object x, Object y) { if (x.isOnGlobalHeap() && !y.isOnGlobalHeap()) { GC.registerReferenceGlobalToLocal(x, y); } }
- 83. Pauseless GC • … and a read barrier //read x.field void checkReadAccess(Object x) { int myTid = getThreadId(); //if this object is thread local and //belongs to another thread, evacuate to global heap if (!x.isOnGlobalHeap() && !x.isInternalTo(myTid)) { x.evaculateToGlobalHeap(); //painful } }
- 84. Pauseless GC • Barriers have to extremely fast, or everything will disappear in overhead
- 85. Pauseless GC • Barriers have to extremely fast, or everything will disappear in overhead
- 86. Pauseless GC • We can, because we own and implemented the thread system
- 87. OS/JVM/Hardware improvements • Threading • Locking • Na,ve memory usage • Virtual address memory usage/exhaus,on • Stack overflows • Page protec,on
- 88. OS/JVM/Hardware improvements • Trap on overflow • Read barriers • Performance counters – IP samples – Cache misses – Userland, please • …
- 89. So?
- 90. Conclusion • It doesn’t hurt to know what’s inside your execu,on environment • In the future – the distance between hardware, OS and run,me will decrease or disappear altogether.
- 91. Q & A?