Paruj Ratanaworabhan Kasetsart University, Thailand
Ben Livshits and Ben Zorn Microsoft Research, Redmond
JSMeter: Characterizing the Behavior of JavaScript Web Applications
1
in collaboration with
David Simmons, Corneliu Barsan, and Allen Wirfs-Brock
Why Measure JavaScript?
• Standardized, de facto language for the web
– Support in every browser, much existing code
• Browser and JavaScript performance is important
– Are current JavaScript benchmarks representative?
– Limited understanding of JavaScript behavior in real sites
• Who cares?
– Users, web application developers, JavaScript engine developers
3
4
ZDNet 29 May 2008
ghacks.net Dec. 2008
Browser
Wars!
Artificial Benchmarks versus Real World Sites
5
Maps Maps
7 V8 programs:
• richards
• deltablue
• crypto
• raytrace
• earley-boyer
• regexp
• splay
8 SunSpider programs:
• 3-draytrace
• access-nbody
• bitops-nsieve
• controlflow
• crypto-aes
• date-xparb
• math-cordic
• string-tagcloud
JSMeter 11 real sites:
Goals of JSMeter Project
• Instrument JavaScript execution and measure behavior
• Compare behavior of JavaScript benchmarks against real sites
• Consider how benchmarks can mislead design decisions
6
How We Measured JavaScript
\ie\jscript\*.cpp
Source-level instrumentation
custom jscript.dll
custom trace files website visits
Offline analyzers
custom trace files
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Constant
Other Str Ops
Concat Op
Visiting the Real Sites
• Getting past page load performance
• Attempted to use each site in “normal” way:
7
amazon Search a book, add to shopping cart, sign in, and sign out bing Type in a search query and also look for images and news bingmap Search for a direction from one city to another cnn Read front page news ebay Search for a notebook, bid, sing in, and sign out economist Read front page news, view comments facebook Log in, visit a friend pages, browse through photos and comments gmail Sign in, check inbox, delete a mail, and sign out google Type in a search query and also look for images and news googlemap Search for a direction from one city to another hotmail Sign in, check inbox, delete a mail, and sign out
Understanding JavaScript Behavior
Code
Events
JavaScript
Objects
8
Code Behavior
Code
Events
JavaScript
Objects
9
• Function size
• Instructions/call
• Code locality
• Instruction mix
Total Bytes of JavaScript Source
10
0
500
1000
1500
2000
2500
am
azon
bin
g
bin
gm
ap
cnn
ebay
econom
ist
gm
ail
map
hotm
ail
richard
s
delt
ablu
e
cry
pto
raytr
ace
earl
ey
regexp
spla
y
3d-ra
ytr
ace
access-nbody
bit
ops-nsie
ve
contr
olf
low
cry
pto
-aes
date
-xparb
math
-cord
ic
regexp-dna
str
ing-ta
gclo
ud
Sourc
e s
ize (kilobyte
s)
Real Sites V8 SunSpider
Code|Objects|Events
Static Unique Functions Executed
12
0
500
1,000
1,500
2,000
2,500
3,000
3,500
4,000am
azon
bin
g
bin
gm
ap
cnn
ebay
econom
ist
gm
ail
map
hotm
ail
richard
s
delt
ablu
e
cry
pto
raytr
ace
earl
ey
regexp
spla
y
3d-ra
ytr
ace
access-nbody
bit
ops-nsie
ve
contr
olf
low
cry
pto
-aes
date
-xparb
math
-cord
ic
regexp-dna
str
ing-ta
gclo
ud
Uniq
ue F
uncti
ons E
xecute
d
Real Sites V8 SunSpider
Code|Objects|Events
Bytecodes / Call
15
0.00
50.00
100.00
150.00
200.00
250.00
300.00
am
azon
bin
g
bin
gm
ap
cnn
ebay
econom
ist
gm
ail
map
hotm
ail
richard
s
delt
ablu
e
cry
pto
raytr
ace
earl
ey
regexp
spla
y
3d-ra
ytr
ace
access-nbody
bit
ops-nsie
ve
contr
olf
low
cry
pto
-aes
date
-xparb
math
-cord
ic
regexp-dna
str
ing-ta
gclo
ud
Byte
codes /
Call
function(a,b) { var i=0,elem,pos=a.length; if(D.browser.msie) { while(elem=b[i++]) if(elem.nodeType!=8) a[pos++]=elem; } else while(elem=b[i++]) a[pos++]=elem; return a }
Real Sites V8 SunSpider
Code|Objects|Events
Fraction of Code Executed
16
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
am
azon
bin
g
bin
gm
ap
cnn
ebay
econom
ist
gm
ail
map
hotm
ail
richard
sdelt
ablu
e
cry
pto
raytr
ace
earl
ey
regexp
spla
y
3d-ra
ytr
ace
access-nbody
bit
ops-nsie
ve
contr
olf
low
cry
pto
-aes
date
-xparb
math
-cord
ic
regexp-dna
str
ing-ta
gclo
ud
Fra
cti
on o
f C
ode E
xecute
d
Most code
not executed
Real Sites V8 SunSpider
Code|Objects|Events
Object Allocation Behavior
Code
Events
JavaScript
Objects
17
• Allocation by types
•Live heap composition
• Lifetime distribution
Total Bytes Allocated
18
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
am
azon
bin
g
bin
gm
ap
cnn
ebay
econom
ist
gm
ail
map
hotm
ail
richard
s
delt
ablu
e
cry
pto
raytr
ace
earl
ey
regexp
spla
y
3d-ra
ytr
ace
access-nbody
bit
ops-nsie
ve
contr
olf
low
cry
pto
-aes
date
-xparb
math
-cord
ic
regexp-dna
str
ing-ta
gclo
ud
Tota
l heap
data
(kilobyte
s)
Real Sites V8 SunSpider
Code|Objects|Events
Heap Data by Type
19
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
amaz
on
bin
gb
ingm
ap cnn
ebay
eco
no
mis
tfa
ceb
oo
kgm
ail
goo
gle
goo
glem
aph
otm
ail
rich
ard
sd
elta
blu
ecr
ypto
rayt
race
earl
eyre
gexp
spla
yV
8 a
ggra
gate
3d
-ray
trac
eac
cess
-nb
od
yb
ito
ps-
nsi
eve
con
tro
lflo
wcr
ypto
-aes
dat
e-xp
arb
mat
h-c
ord
icre
gexp
-dn
ast
rin
g-ta
gclo
ud
aggr
egat
e
He
ap D
ata
by
Typ
e
Objects
Enumarator
RegExp
Math
Error
Date
Native Func
String
Arrays
Script Func
Real Sites V8 SunSpider
Many functions
Rest are strings
Few benchmarks
allocate much data
Code|Objects|Events
Live Heap Over Time (gmail)
20
Functions grow
steadily
Code|Objects|Events
GC reduces
size of heap
Objects grow
steadily too
Live Heap over Time (ebay)
21
Heaps
repeatedly
created,
discarded
Heap contains
mostly functions
Code|Objects|Events
Heap drops to 0
on page load
2 Search Websites, 2 Architectures
Bing Google
Code|Objects|Events
22
You stay on the same page
during your entire visit
Code loaded once
Heap is bigger
Every transition loads a
new page
Code loaded repeatedly
Heap is smaller
Event Handlers in JavaScript
Code
Events
JavaScript
Objects
26
• Number of events • Sizes of handlers
Event-driven Programming Model
• Single-threaded, non-preemptive event handlers
• Example handlers: onabort, onclick, etc.
• Very different from batch processing of benchmarks
• Handler responsiveness critical to user experience
27
Code|Objects|Events
Total Events Handled
28
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000T
ota
l Events
Handle
d
Real Sites V8
Almost no events
Code|Objects|Events
Median Bytecodes / Event Handled
30
0
50
100
150
200
250
300
350
400
450
500M
edia
n B
yte
codes /
Event
Hand
led
Code|Objects|Events
506 2137
Sure, this is all good, but…
• Everyone knows benchmarks are unrepresentative
• How much difference does it make, anyway?
• Wouldn’t any benchmarks have similar issues?
33
Cold-code Experiment
• Observation
– Real web apps have lots of code (much of it cold)
– Benchmarks do not
• Question: What happens if the benchmarks have more code?
– We added extra, unused to code to 7 SunSpider benchmarks
– We measured the impact on the benchmark performance
34
Performance Impact of Cold Code
35
0
100
200
300
400
500
600
700
800
900
Tim
e (
mse
c)
0K
200K
400K
800K
1M
2M
0
100
200
300
400
500
600
700
800
900
Tim
e (
mse
c)
0K
200K
400K
800K
1M
2M
Chrome
3.0.195.38
IE 8
8.0.601.18865 Cold code has
non-uniform impact
on execution time
Cold code makes
SunSpider on Chrome
up to 4.5x slower
Impact of Benchmarks
• What gets emphasis
– Making tight loops fast
– Optimizing small amounts of code
• Important issues ignored
– Garbage collection (especially of strings)
– Managing large amounts of code
– Optimizing event handling
– Considering JavaScript context between page loads
36
Conclusions • JSMeter is an instrumentation framework
– Used to measure and compare JavaScript applications
– High-level views of behavior promote understanding
• Benchmarks differ significantly from real sites
– Misleads designers, skews implementations
• Next steps
– Develop and promote better benchmarks
– Design and evaluate better JavaScript runtimes
– Promote better performance tools for JavaScript developers
37
Additional Resources
• Project: http://research.microsoft.com/en-us/projects/jsmeter/
• Video: Project JSMeter: JavaScript Performance Analysis in the Real World" - MSDN Channel 9 interview with Erik Meier, Ben Livshits, and Ben Zorn
• Paper:
– “JSMeter: Comparing the Behavior of JavaScript Benchmarks with Real Web Applications”, Paruj Ratanaworabhan, Benjamin Livshits and Benjamin G. Zorn, USENIX 2010 Conference on Web Application Development (WebApps’10), June 2010.
• Additional measurements: – "JSMeter: Characterizing Real-World Behavior of JavaScript Programs", Paruj
Ratanaworabhan, Benjamin Livshits, David Simmons, and Benjamin Zorn, MSR-TR-2009-173, December 2009 (49 pages), November 2009.
38
Additional Slides
39
Hot Function Distribution
40
Real Sites V8 Benchmarks
80% of time in < 10 functions 80% of time in 100+ functions
80% 80%
Code|Objects|Events
Opcode Distribution
41 Green = SunSpider | Blue= Real Web Apps | Red = V8
Real Apps
Outliers
Code|Objects|Events
Object Type Distribution
42
Real Apps
Code|Objects|Events
economist
is an outlier
(arrays)
Green = SunSpider | Blue= Real Web Apps | Red = V8
Distribution of Time in Handlers
43
Code|Objects|Events
Related Work • JavaScript
– Richards, Lebresne, Burg, and Vitek (PLDI’10)
– Draw similar conclusions
• Java
– Doufour et al. (OOPSLA’03), Dieckmann and U. Hölzle (ECOOP’99)
• Other languages
– C++: Calder et al. (JPL’95)
– Interpreted languages: Romer et al. (ASPLOS’96)
44
