Upgrade to Pro — share decks privately, control downloads, hide ads and more …

The Morning Paper: foundations & frontiers

The Morning Paper: foundations & frontiers

Talk from Craft-Conf 2017

Adrian Colyer

April 28, 2017
Tweet

More Decks by Adrian Colyer

Other Decks in Research

Transcript

  1. The Morning Paper:
    foundations and frontiers
    Adrian Colyer
    @adriancolyer

    View full-size slide

  2. blog.acolyer.org
    550
    Foundations
    Frontiers

    View full-size slide

  3. Brain
    storm
    01
    02
    05
    04
    rainstorm
    03
    5 Reasons to <3 Papers
    Thinking
    tools
    Raise
    Expectations
    Applied
    Lessons Order of
    magnitude
    breakthroughs
    Heads-up
    3

    View full-size slide

  4. 4
    01
    02
    03
    04
    05
    Software development
    Distributed Systems & Big Data
    Infrastructure implications
    Security
    ML & DL

    View full-size slide

  5. Software
    Development
    5

    View full-size slide

  6. $PRGXOHLVDXQLWRIZRUNDVVLJQPHQW
    6KRUWHQGHYHORSPHQW
    WLPH
    ,PSURYHV\VWHP
    IOH[LELOLW\
    ,PSURYH
    XQGHUVWDQGDELOLW\!
    EHWWHURYHUDOOGHVLJQ
    Ɣ ,QGHSHQGHQW
    GHSOR\PHQW
    Ɣ )LQHJUDLQHGVFDOLQJ
    Ɣ )DXOWLVRODWLRQ

    View full-size slide

  7. &RS\ULJKW0D[LP3RSRY5)6WRFN3KRWR
    ŸEYVVWWVTeZgV_Vdd`W
    R^`Uf]RcZkReZ`_Zd
    UVaV_UV_efa`_eYV
    TcZeVcZRfdVUZ_
    UZgZUZ_XeYVdjdeV^
    Z_e`^`Uf]VdŹ

    View full-size slide

  8. Circa 1979 (& 2016!)
    &RPPRQ3UREOHPV
    :HZHUHEHKLQGVFKHGXOHDQG
    ZDQWHGWRGHOLYHUDQHDUO\
    UHOHDVHEXWIRXQGWKDWZH
    FRXOGQşWVXEVHWWKHV\VWHP
    :HZDQWHGWRDGGDVLPSOH
    IHDWXUHEXWIRXQGLWZRXOG
    KDYHUHTXLUHGUHZULWLQJDOORU
    PRVWRIWKHFXUUHQWFRGH
    :HZDQWHGWRVLPSOLI\WKH
    V\VWHPE\UHPRYLQJVRPH
    IHDWXUHEXWWDNLQJDGYDQWDJH
    RILWPHDQWUHZULWLQJODUJH
    VHFWLRQVRIWKHFRGH
    :HZDQWHGDFXVWRP
    GHSOR\PHQWHJLQGHYRUWHVW
    HQYLURQPHQWVEXWWKHV\VWHP
    ZDVQşWIOH[LEOHHQRXJK

    View full-size slide

  9. J>;HKB;I0
    C_Yhei[hl_Y[7_iWbbem[Zjeki[c_Yhei[hl_Y[8_\\0
    Ɣ 7_i[ii[dj_Wbboi_cfb[hX[YWki[_jki[i8
    Ɣ 8_idejikXijWdj_Wbboceh[Yecfb[nX[YWki[_j_idejWbbem[Zjeki[7
    Ɣ J^[h[_iWki[\kbikXi[jYedjW_d_d]8WdZdej7
    Ɣ J^[h[_ideYedY[_lWXb[ki[\kbikXi[jYedjW_d_d]7Xkjdej8
    7dZe\Yekhi["_jZe[idej_djheZkY[WdoYoYb[i_djej^[Z[f[dZ[dYo]hWf^

    View full-size slide

  10. Ÿ2WeVcViR^Z_Z_XYf_UcVUd`WVcc`cac`_V
    5CDaRTVd`gVcU`kV_d`W`aV_d`fcTVR_U
    T`^^VcTZR]ac`[VTedhVYRgV`SdVcgVUeYRe
    eYVcVRcV[fdeRWVhUZdeZ_TeejaVd`W
    RcTYZeVTefcVZddfVdR_UeYVdV`TTfc`gVcR_U
    `gVcRXRZ_žŹ

    View full-size slide

  11. %) %XJ)UHTXHQF\%& %XJFKXUQ&) &KDQJH)UHTXHQF\&& &KDQJH&KXUQ
    9`h^fTY
    h`cdVW`c
    RcTYZeVTefcV
    Y`eda`ed0

    View full-size slide

  12. C7?DIEKH9;IE<C7?DJ;D7D9;
    9EIJI0
    '$ KdijWXb[_dj[h\WY[
    ($ ?cfb_Y_jYheii#ceZkb[Z[f[dZ[dYo
    )$ Kd^[Wbj^o_dj[h\WY[_d^[h_jWdY[
    ^_[hWhY^o
    *$ 9heii#ceZkb[YoYb[
    +$ 9heii#fWYaW][YoYb[

    View full-size slide

  13. 7KHGDWDVD\V
    7KHWZRPRVWLPSRUWDQWDUHDVWRSD\DWWHQWLRQWRDUH
    Ɣ WKHLQWHUIDFHVRIWKHPRGXOHVDQGKRZZHOOWKH\KLGH
    LQIRUPDWLRQVRWKDWFKDQJHVFDQEHPDGHZLWKRXW
    FDVFDGHVDQG
    Ɣ WKHXVHVVWUXFWXUHRIWKHV\VWHP

    View full-size slide

  14. ,GHQWLI\LQJDQGTXDQWLI\LQJDUFKLWHFWXUDOGHEW
    Ɣ $UFKLWHFWXUDOGHEWVFRQVXPHRIWKHWRWDOSURMHFWPDLQWHQDQFHHIIRUWLQ
    SURMHFWVVWXGLHG
    Ɣ 7KHWRSILYHPRGXODULW\GHEWVDORQHFRQVXPHRIWKHWRWDOHIIRUW
    Ɣ 0RGXODULW\YLRODWLRQLVWKHPRVWFRPPRQDQGH[SHQVLYHGHEWRYHUDOOLW
    DFFRXQWVIRURIWKHWRWDOHIIRUWLQ+%DVH
    Ɣ 7RSGHEWVRQO\LQYROYHDVPDOOQXPEHURIILOHVPRGXOHVEXWFRQVXPHDODUJH
    DPRXQWRIWKHWRWDOSURMHFWHIIRUW
    Ɣ $ERXWKDOIRIDOODUFKLWHFWXUDOGHEWVDFFXPXODWHLQWHUHVWDWDFRQVWDQWUDWH

    View full-size slide

  15. Ÿ2]^`deR]]TReRdec`aYZTWRZ]fcVd%)Z_
    e`eR]ų*#RcVeYVcVdf]e`W
    Z_T`ccVTeYR_U]Z_X`W_`_WReR]Vcc`cd
    Via]ZTZe]jdZX_R]]VUZ_d`WehRcVŹ

    View full-size slide

  16. Ÿ5VdaZeVR]]eYVVWW`ced`WgR]ZUReZ`_
    cVgZVhR_UeVdeZ_XT`_WZXfcReZ`_
    Vcc`cddeZ]]TRfdV^R_jYZXYZ^aRTe
    Z_TZUV_ed`We`URjŶd:_eVc_VeR_UT]`fU
    djdeV^dŹ

    View full-size slide

  17. Distributed Systems
    and Big Data
    20

    View full-size slide

  18. Frank McSherry
    Scalability - but at what COST?
    21

    View full-size slide

  19. But you have BIG Data!
    23
    Zipf Distribution
    ³:RUNLQJVHWVDUH
    =LSIGLVWULEXWHG:HFDQ
    WKHUHIRUHVWRUHLQPHPRU\DOO
    EXWWKHYHU\ODUJHVW
    GDWDVHWV´

    View full-size slide

  20. Musketeer
    24
    One for all?

    View full-size slide

  21. Approx Hadoop
    25
    32x!

    View full-size slide

  22. HopFS - FAST’17
    26

    View full-size slide

  23. Redundancy does not imply fault tolerance -
    FAST’17
    27
    ŸRdZ_X]VWZ]VdjdeV^WRf]eTR_
    Z_UfTVTReRdec`aYZT`feT`^VdZ_
    ^`de^`UVc_UZdecZSfeVUde`cRXV
    djdeV^dUReR]`ddT`ccfaeZ`_
    f_RgRZ]RSZ]ZejR_UZ_d`^V
    TRdVdeYVdacVRU`WT`ccfaeZ`_
    e``eYVcZ_eRTecVa]ZTRdŹ

    View full-size slide

  24. Infrastructure
    implications
    28

    View full-size slide

  25. +XPDQ
    FRPSXWHUV
    DW'U\GHQE\1$&$1$6$
    'U\GHQ)OLJKW5HVHDUFK&HQWHU
    3KRWR&ROOHFWLRQ
    KWWSZZZGIUFQDVDJRY*DOOHU\3KRWR3ODFHV+7
    0/(KWPO/LFHQVHGXQGHU3XEOLF'RPDLQYLD
    &RPPRQV
    KWWSVFRPPRQVZLNLPHGLDRUJZLNL)LOH+XPDQBFR
    PSXWHUVBB'U\GHQMSJPHGLD)LOH+XPDQBFRPSXW
    HUVBB'U\GHQMSJ

    View full-size slide

  26. Computing on a Human Scale
    30
    10ns
    70ns
    10ms
    10s
    1:10s
    116d
    5HJLVWHUV
    //
    )LOHRQ
    GHVN
    0DLQ
    PHPRU\
    2IILFHILOLQJ
    FDELQHW
    +''
    7ULSWRWKH
    ZDUHKRXVH

    View full-size slide

  27. Compute
    HTM
    Persistent Memory NI
    FPGA
    GPUs
    Memory
    NVDIMMs
    Persistent Memory
    Networking
    100GbE
    RDMA
    Storage
    NVMe
    Next-gen NVM
    Next Generation Hardware
    All Change Please
    31

    View full-size slide

  28. 2-10m
    Computing on a Human Scale
    32
    10s
    1:10s
    116d
    File on
    desk
    Office filing
    cabinet
    Trip to the
    warehouse
    4x capacity
    fireproof local
    filing cabinets
    23-40m
    Phone
    another office
    (RDMA)
    3h20m Next-gen
    warehouse

    View full-size slide

  29. The New ~Numbers Everyone Should Know
    33
    Latency Bandwidth Capacity/IOPS
    Register 0.25ns
    L1 cache 1ns
    L2 cache 3ns 8MB
    L3 cache 11ns 45MB
    DRAM 62ns 120GBs 6TB - 4 socket
    NVRAM’ DIMM 620ns 60GBs 24TB - 4 socket
    1-sided RDMA in Data Center 1.4us 100GbE ~700K IOPS
    RPC in Data Center 2.4us 100GbE ~400K IOPS
    NVRAM’ NVMe 12us 6GBs 16TB/disk,~2M/600K
    NVRAM’ NVMf 90us 5GBs 16TB/disk, ~700/600K

    View full-size slide

  30. No Compromises - FaRM
    34
    TPC-C (90 nodes)
    4.5M tps
    99%ile
    1.9ms
    KV (per node)
    6.3M qps
    at peak throughput
    41μs

    View full-size slide

  31. No Compromises
    35
    ³7KLVSDSHUGHPRQVWUDWHVWKDWQHZVRIWZDUHLQPRGHUQ
    GDWDFHQWHUVFDQHOLPLQDWHWKHQHHGWRFRPSURPLVH,W
    GHVFULEHVWKHWUDQVDFWLRQUHSOLFDWLRQDQGUHFRYHU\
    SURWRFROVLQ)D50DPDLQPHPRU\GLVWULEXWHGFRPSXWLQJ
    SODWIRUP)D50SURYLGHVGLVWULEXWHG$&,'WUDQVDFWLRQV
    ZLWKVWULFWVHULDOL]DELOLW\KLJKDYDLODELOLW\KLJK
    WKURXJKSXWDQGORZODWHQF\7KHVHSURWRFROVZHUH
    GHVLJQHGIURPILUVWSULQFLSOHVWROHYHUDJHWZRKDUGZDUH
    WUHQGVDSSHDULQJLQGDWDFHQWHUVIDVWFRPPRGLW\
    QHWZRUNVZLWK5'0$DQGDQLQH[SHQVLYHDSSURDFKWR
    SURYLGLQJQRQYRODWLOH'5$0´

    View full-size slide

  32. DrTM
    The Doctor will see you now
    36
    5.5M tps on TPC-C
    6-node cluster.

    View full-size slide

  33. Making smart contracts smarter
    CCS ‘16
    38
    19,366
    contracts
    $30M
    USD
    8,833
    vulnerable
    27.9% 15.7% 340 83

    (UURU
    H[FHSWLRQ
    KDQGOLQJ

    7UDQVDFWLRQ
    RUGHULQJ
    5HHQWUDQF\
    KDQGOLQJ
    7LPHVWDPS
    RUGHULQJ

    View full-size slide

  34. OSDI ‘16
    Scone: Secure Linux containers with Intel SGX
    39

    View full-size slide

  35. NDSS ‘17
    Thou shalt not depend on me
    40
    37% vulnerable
    jQuery -> 36.7%,
    Angular -> 40.1%

    View full-size slide

  36. lessons from Google
    Machine Learning Systems
    42
    Feature
    Management
    Visualisation
    Relative Metrics
    Systematic Bias
    Correction
    Alerts on action
    Thresholds
    01
    02
    03
    04
    05

    View full-size slide

  37. ICLR 2015
    Explaining and harnessing adversarial examples
    43

    View full-size slide

  38. CVPR ‘15
    Deep neural networks are easily fooled
    44

    View full-size slide

  39. Wrapping Up
    45

    View full-size slide

  40. Brain
    storm
    01
    02
    05
    04
    rainstorm
    03
    5 Reasons to <3 Papers
    Thinking
    tools
    Raise
    Expectations
    Applied
    Lessons Order of
    magnitude
    breakthroughs
    Heads-up
    46

    View full-size slide

  41. Don’t just take my word for it...
    47
    HYV_:eR]\e`cVdVRcTYVcdhYV_:eR]\
    e`aV`a]VhR_eZ_Xe`V_XRXVZ_
    V_ecVacV_VfcdYZa:eV]]eYV^eYReZWj`f
    cVRUcVdVRcTYaRaVcdT`_dZdeV_e]jZW
    j`fdVcZ`fd]jdefUjYR]WRU`kV_aRaVcd
    RhVV\R_Uj`fU`eYReW`ceh`jVRcd
    RWeVceY`dVeh`jVRcdj`fhZ]]YRgV
    ]VRc_VUR]`eEYZdZdRWR_eRdeZT
    Z_gVde^V_eZ_j`fc`h_]`_XeVc^
    UVgV]`a^V_e
    $QGUHZ1J
    ³,QVLGHWKHPLQGWKDWEXLOW*RRJOH%UDLQ´

    KWWSZZZKXIILQJWRQSRVWFRPDX
    DQGUHZQJBQBKWPO

    View full-size slide

  42. Don’t just take my word for it...
    48
    :U`_Ŷe\_`hY`heYVYf^R_ScRZ_
    h`c\dSfeZeŶdR]^`de^RXZTR]hYV_
    j`fcVRUV_`fXY`ceR]\e`V_`fXY
    ViaVcedhYV_j`fYRgVV_`fXYZ_afed
    _VhZUVRddeRceRaaVRcZ_X
    $QGUHZ1J
    ³,QVLGHWKHPLQGWKDWEXLOW*RRJOH%UDLQ´
    KWWSZZZKXIILQJWRQSRVWFRPDXDQGUHZQJBQBKWPO

    View full-size slide

  43. A new paper every weekday
    Published at http://blog.acolyer.org.
    01
    Delivered Straight to your inbox
    If you prefer email-based subscription to read at
    your leisure.
    02
    Announced on Twitter
    I’m @adriancolyer.
    03
    Go to a Papers We Love Meetup
    A repository of academic computer science papers
    and a community who loves reading them.
    04
    Share what you learn
    Anyone can take part in the great conversation.
    05

    View full-size slide

  44. THANK YOU !
    @adriancolyer

    View full-size slide