28. Where they fit in the picture
13(N&!/#$%&'!! "<NJ!/#$%&'!!
E2'3/#'8C;#:X<2I!
K.%#N!.')&'C;#$&)! F*BC;#$&)!
B&'Jf:#J.(! 1%#%&!/#:+3(&!'&2N3:#J.(!
@.(C7.J(! n.J(!
dRB!
BN.<)73&P$!FGMM>!*.'%.>!*.'%<#N>!@.7]!U!FGMM!
32. Non-voting
K^d!.-!KMh$! K^d!.-!KFh$!
'&#)!`!P'3%&$&%! '&#)!`!P'3%&$&%!
A[&:<J.(!
K'#($#:J.(!KM!
9M!
A[&:<J.(! n#N3)#J.(`B.//3%! n#N3)#J.(`V;.'%!
K'#($#:J.(!KF! KM! KF!
9F!
n#N3)#J.(`B.//3%! n#N3)#J.(`V;.'%!
KM! KF!
9W!
+ only validation executed at all replicas:
high scalability with write intensive workloads
- need to send also readset: often very large!
!
BN.<)73&P$!FGMM>!*.'%.>!*.'%<#N>!@.7]!U!FGMM!
45. Chasing the optimum!
("
!#'"
!"#$%&'()*+,-#"./-0.1+
!#&" *+,"
-.,"
!#%" +,"
/0123,-"
!#$"
/4"
!"
(" "()!!!"" "(!!)!!!""
2)%*+3)1+3'()+
Fig. 5. Normalized throughput of the adaptive and non-adaptive protocols (Bank benchmark).
BN.<)73&P$!FGMM>!*.'%.>!*.'%<#N>!@.7]!U!FGMM!
transactions, thus allowing us to focus on the performance of the transactions that re-
46. !and beating it!
(#%"
(#$"
!"#$%&'()*+,-#"./-0.1+
("
)*"
!#'" +,"
!#&" -*."
+/01-*."
!#%"
!#$"
!"
Fig. 8. Normalized throughput of the adaptive and VC protocols. NVC and BFC not reported as
BN.<)73&P$!FGMM>!*.'%.>!*.'%<#N>!@.7]!U!FGMM!
they caused the collapse of the GCS layer. (STMBench7 benchmark)
52. 2PC-based replication
• Transactions executed at all nodes w/o coordination till
commit time
• Acquire atomically locks at all nodes using two phase
commit (2PC):
• 2PC materializes conflicts among concurrent remote transactions
generating:
DISTRIBUTED DEADLOCKS
+ good scalability at low conflict
- thrashes at high conflict
!
BN.<)73&P$!FGMM>!*.'%.>!*.'%<#N>!@.7]!U!FGMM!
65. *#.N.!9./#(.!#()!"#o&.!D&.(&|!
@0)A-.+78761B5.(!8761871C*.5)1D=,0=1B=*5,(52.1E=*.*(*)213851F75)9G(5)1
H*,0))876157,1I087A*=(0/07.1J05=78761
?AAA!?(%&'(#J.(#N!B.(-&'&(:&!.(!B./2<J(>!@&%P.'X3(!#()!
B.//<(3:#J.($!E?B@BwMFI>!T#(]!FGMF!
!
BATCHING IN TOTAL ORDER
BROADCAST PROTOCOLS
BN.<)73&P$!FGMM>!*.'%.>!*.'%<#N>!@.7]!U!FGMM!
68. Analytical model
• 6$3(!_<&<3(!%+&.'8!#'</&(%$!P&!:#(!
)&%&'/3(&!%+&!.2J/#N!;#%:+3(!J/&>!;}>!#$!#!
-<(:J.(!.-!%+&!:<''&(%!N.#)>!/Q1
ueue: linearly and has a vertical asymptote at m = m∗ .
2
1, if m Tadd σ 2 4σ 2 +2Tadd σ 4
(1)
2 + 1
2 2
). We b∗ (m) = 2m−σ−2mTadd σ (7)
σ−2mT σ+
,
b, m)
2(σ+2m(Tadd σ−1))2
add (2σT −1)2 (1+2σT
)σ 2
add add
Tadd σ 2
2
4σ 2 +2Tadd σ 4
if 2 + 1
2 2 m m∗
(2)
BN.)73P$!FGMM!*.'%.!*.'%#N!@.7]!U!FGMM!
l rate B. Determining the Model’s Parameters
using
69. 60
Optimal B
Avg Msg. Arriv
4000
40
20 Model Accuracy 2000
0
0 2000 4000 6000 8000 10000 12000 14000 0 2
Average Msg. Arrival Rate (msgs/sec)
100 Figure 4. Traffi
Exaustive Manual Tuning
Analytical Model
.)N!()'$J/#%$!
.2J/#N!;#%:+3(!7#N!
Optimal Batching Value
#%!0/)3/5!N.#)e!
! 100
Latency (msec)
10 70G#)F.
;5.(!8761
10
+7,0=02G/5G*71(5+2021
292.0/1872.5;8)8.91
1
1
msgs/sec
0 2000 4000 6000 8000 10000 12000 14000 10000
Average Msg. Arrival Rate (msgs/sec)
5000
BN.)73P$!FGMM!*.'%.!*.'%#N!@.7]!U!FGMM!
Figure 3. Validating the accuracy of the analytical model. 0
16 1
70. 2000
A 0
0 2 4 6 8 10 12 14 16 18 20 22 24
Peak period analysis
Hour of the day (3 Sept. 2010)
Figure 4. Traffic at the FenixEDU system (Sept. 3 2010)
9#/2C2!`!'#/2C).P(!
%'#($3J.(!%+'.+!%+!
100 02'.;N/#J:5!#'#$Q!
Latency (msec)
C '#/2C2!3$!$c:3(%N8!
$+.'%Q!
10 ! $8$%/!0$%'N$5!;%!
':.7'$!
1 C '#/2C).P(!3$!N.('Q!
! !!
msgs/sec
10000
5000
0
16 17 18 19 20 21 22
Hour of the day
Figure 5. Self-tuning based on analytical model
BN.)73P$!FGMM!*.'%.!*.'%#N!@.7]!U!FGMM!
71. What about a pure ML approach?
• *'.;N/Q!
– D!%:+(3_$!()!%.![2N.'!)3l'(%!$.NJ.($!
E;#%:+3(!7#N$I!%.!3)(J-8!.2J/#N!.(Q!
• N.P!N.#)Q!$N$$!#))3J.(#N!N#%(:8!
!
• /)3/C+3+!N.#)Q!3($c:3(%!;#%:+3(!7#N$!N#)!
7'8!'#23)N8!%.!3($%#;3N3%8!#()!%+'#$+3(!
!
BN.)73P$!FGMM!*.'%.!*.'%#N!@.7]!U!FGMM!
72. Combining the two approaches
M] ?(3J#N34!Dh$!X(.PN)!P3%+!%+!2')3:J.($!
.-!%+!#(#N8J:#N!/.)NQ!
– '):!-'_(:8!.-!.;73.$N8!P'.(![2N.'#J.($!
F] D%!D!2)#%!%+!3(3J#N!'P#')!7#N$Q!
– :.'':%!/.)Nh$!2')3:J.(!''.'$![2N.3J(!
-);#:X!-'./!%+!$8$%/!
BN.)73P$!FGMM!*.'%.!*.'%#N!@.7]!U!FGMM!
73. Combining the two approaches
lytical Model
ry ex- 100
Model+RL
ent of
Latency (msec)
ired to
whole 10
ed the
l Core 1
e time
der of
msgs/sec
10000
5000
hough 0
etailed 16 17 18 19 20 21 22
ith the Hour of the day
above
Figure 6. Self-tuning combining RL and analytical model
erage) BN.)73P$!FGMM!*.'%.!*.'%#N!@.7]!U!FGMM!
ntially
generic
the reward of each possible decision, and lowers it as more