Everything I Ever Learned About JVM Performance Tuning @Twitter

Attila Szegedi, Software Engineer
@asz

1

Everything I ever
learned about JVM
performance tuning
@twitter

2

Everything More
than I ever wanted
to learned about
JVM performance
tuning
@twitter
3

• Memory tuning
• CPU usage tuning
• Lock contention tuning
• I/O tuning

4

Twitter’s biggest enemy

5

Latency

5

Latency

CC licensed image from http://www.flickr.com/photos/dunechaser/
213255210/ 5

Latency contributors

• By far the biggest contributor is garbage collector
• others are, in no particular order:
• in-process locking and thread scheduling,
• I/O,
• application algorithmic inefﬁciencies.
6

Areas of performance
tuning

• Memory tuning
• Lock contention tuning
• CPU usage tuning
• I/O tuning
7

Areas of memory
performance tuning

• Memory footprint tuning
• Allocation rate tuning
• Garbage collection tuning

8

Memory footprint tuning

• So you got an OutOfMemoryError…
• Maybe you just have too much data!
• Maybe your data representation is fat!
• You can also have a genuine memory leak…
9

Too much data

• Run with -verbosegc
• Observe numbers in “Full GC” messages secs]
[Full GC $before->$after($total), $time

• Can you give the JVM more memory?
• Do you need all that data in memory? Consider
using:
• a LRU cache, or…
• soft references* 10

Fat data

• Can be a problem when you want to do wacky
things, like
• load the full Twitter social graph in a single
JVM
• load all user metadata in a single JVM
• Slimming internal data representation works at
these economies of scale
11

Fat data: object header

• JVM object header is normally two machine
words.
• That’s 16 bytes, or 128 bits on a 64-bit JVM!
• new java.lang.Object() takes 16 bytes.
• new byte[0] takes 24 bytes.

12

Fat data: padding

class A {
byte x;
}
class B extends A {
byte y;
}

• new A() takes 24 bytes.

• new B() takes 32 bytes.
13

Fat data: no inline structs

class C {
Object obj = new Object();
}

• new C() takes 40 bytes.
• similarly, no inline array elements.
14

Slimming taken to
extreme
• A research project had to load the full follower
graph in memory
• Each vertex’s edges ended up being represented
as int arrays
• If it grows further, we can consider variable-
length differential encoding in a byte array

15

Compressed object
pointers

• Pointers become 4 bytes long
• Usable below 32 GB of max heap size
• Automatically used below 30 GB of max heap

16

Compressed object
pointers
Uncompressed Compressed 32-bit

Pointer 8 4 4

Object header 16 12* 8

Array header 24 16 12

Superclass pad 8 4 4

* Object can have 4 bytes of ﬁelds and still only take up 16 bytes
17

Avoid instances of
primitive wrappers

• Hard won experience with Scala 2.7.7:
• a Seq[Int] stores java.lang.Integer
• an Array[Int] stores int
• ﬁrst needs (24 + 32 * length) bytes
• second needs (24 + 4 * length) bytes
18

Avoid instances of
primitive wrappers

• This was ﬁxed in Scala 2.8, but it shows that:
• you often don’t know the performance
characteristics of your libraries,
• and won’t ever know them until you run your
application under a proﬁler.

19

Map footprints

• Guava MapMaker.makeMap() takes 2272 bytes!

• MapMaker.concurrencyLevel(1).makeMap()
takes 352 bytes!
• ConcurrentMap with level 1 makes sense
sometimes (i.e. you don’t want a
ConcurrentModiﬁcationException)

20

Thrift can be heavy

• Thrift generated classes are used to encapsulate a
wire tranfer format.
• Using them as your domain objects: almost never
a good idea.

21

Thrift can be heavy

• Every Thrift class with a primitive ﬁeld has a
java.util.BitSet __isset_bit_vector ﬁeld.

• It adds between 52 and 72 bytes of overhead per
object.

22

Thrift can be heavy

23

Thrift can be heavy

• Thrift does not support 32-bit ﬂoats.
• Coupling domain model with transport:
• resistance to change domain model
• You also miss oportunities for interning and N-to-1
normalization.

24

class Location {
public String city;
public String region;
public String countryCode;
public int metro;
public List<String> placeIds;
public double lat;
public double lon;
public double confidence;

25

class SharedLocation {
public String city;
public String region;
public String countryCode;
public int metro;
public List<String> placeIds;
class UniqueLocation {
private SharedLocation sharedLocation;
public double lat;
public double lon;
public double confidence;

26

Careful with thread locals

• Thread locals stick around.
• Particularly problematic in thread pools with m⨯n
resource association.
• 200 pooled threads using 50 connections: you end
up with 10 000 connection buffers.
• Consider using synchronized objects, or
• just create new objects all the time.
27

Part II:
fighting latency

28

Performance tradeoff

Memory

Time

Convenient, but oversimpliﬁed view.
29

Performance triangle

Memory footprint

Throughput Latency

30


Compactness

Throughput Responsiveness
C ⨯T ⨯ R = a
• Tuning: vary C, T, R for ﬁxed a
• Optimization: increase a 31


• Compactness: inverse of memory footprint
• Responsiveness: longest pause the application will
experience
• Throughput: amount of useful application CPU work
over time
• Can trade one for the other, within limits.
• If you have spare CPU, can be pure win.
32

Responsiveness vs.
throughput

33

Biggest threat to
responsiveness in the JVM
is the garbage collector

34

Memory pools

Eden Survivor Old

Code
Permanent
cache

This is entirely HotSpot speciﬁc!
35

How does young gen
work?
Eden S1 S2 Old

• All new allocation happens in eden.
• It only costs a pointer bump.
• When eden ﬁlls up, stop-the-world copy-collection
into the survivor space.
• Dead objects cost zero to collect.
• Aftr several collections, survivors get tenured into
old generation.
36

Ideal young gen operation

• Big enough to hold more than one set of all
concurrent request-response cycle objects.
• Each survivor space big enough to hold active
request objects + tenuring ones.
• Tenuring threshold such that long-lived objects
tenure fast.

37

Old generation collectors

• Throughput collectors
• -XX:+UseSerialGC

• -XX:+UseParallelGC

• -XX:+UseParallelOldGC

• Low-pause collectors
• -XX:+UseConcMarkSweepGC

• -XX:+UseG1GC (can’t discuss it here)

38

Adaptive sizing policy

• Throughput collectors can automatically tune
themselves:
• -XX:+UseAdaptiveSizePolicy

• -XX:MaxGCPauseMillis=… (i.e. 100)
• -XX:GCTimeRatio=… (i.e. 19)

39

Adaptive sizing policy at
work

40

Choose a collector

• Bulk service: throughput collector, no adaptive sizing
policy.
• Everything else: try throughput collector with
adaptive sizing policy. If it didn’t work, use
concurrent mark-and-sweep (CMS).

41

Always start with tuning
the young generation
• Enable -XX:+PrintGCDetails, -XX:+PrintHeapAtGC,
and -XX:+PrintTenuringDistribution.

• Watch survivor sizes! You’ll need to determine
“desired survivor size”.
• There’s no such thing as a “desired eden size”, mind
you. The bigger, the better, with some
responsiveness caveats.
• Watch the tenuring threshold; might need to tune it
to tenure long lived objects faster.
42

-XX:+PrintHeapAtGC

Heap after GC invocations=7000 (full 87):
par new generation total 4608000K, used 398455K
eden space 4096000K, 0% used
from space 512000K, 77% used
to space 512000K, 0% used
concurrent mark-sweep generation total 3072000K, used 1565157K
concurrent-mark-sweep perm gen total 53256K, used 31889K
}

43

-XX:+PrintTenuringDistribution

Desired survivor size 262144000 bytes, new threshold 4 (max 4)
- age 1: 137474336 bytes, 137474336 total
- age 2: 37725496 bytes, 175199832 total
- age 3: 23551752 bytes, 198751584 total
- age 4: 14772272 bytes, 213523856 total

• Things of interest:
• Number of ages
• Size distribution in ages
• You want strongly declining.
44

Tuning the CMS

• Give your app as much memory as possible.
• CMS is speculative. More memory, less punitive
miscalculations.
• Try using CMS without tuning. Use -verbosegc and
-XX:+PrintGCDetails.

• Didn’t get any “Full GC” messages? You’re done!
• Otherwise, tune the young generation ﬁrst.
45

Tuning the old generation

• Goals:
• Keep the fragmentation low.
• Avoid full GC stops.
• Fortunately, the two goals are not conﬂicting.

46


• Find the minimum and maximum working set size
(observe “Full GC” numbers under stable state and
under load).
• Overprovision the numbers by 25-33%.
• This gives CMS a cushion to concurrently clean
memory as it’s used.

47


• Set -XX:InitiatingOccupancyFraction to
between 80-75, respectively.
• corresponds to overprovisioned heap ratio.
• You can lower initiating occupancy fraction to 0 if
you have CPU to spare.

48

Responsiveness still not
good enough?
• Too many live objects during young gen GC:
• Reduce NewSize, reduce survivor spaces, reduce
tenuring threshold.
• Too many threads:
• Find the minimal concurrency level, or
• split the service into several JVMs.
49

Responsiveness still not
good enough?
• Does the CMS abortable preclean phase, well,
abort?
• It is sensitive to number of objects in the new
generation, so
• go for smaller new generation
• try to reduce the amount of short-lived garbage
your app creates.

50

Part III:
let’s take a break from GC

51

Thread coordination
optimization

• You don’t have to always go for synchronized.
• Synchronization is a read barrier on entry; write
barrier on exit.
• Sometimes you only need a half-barrier; i.e. in a
producer-observer pattern.
• Volatiles can be used as half-barriers.
52

Thread coordination
optimization

• For atomic update of a single value, you only need
Atomic{Integer|Long}.compareAndSet().

• You can use AtomicReference.compareAndSet() for
atomic update of composite values represented by
immutable objects.

53

Fight CMS fragmentation
with slab allocators

• CMS doesn’t compact, so it’s prone to fragmentation,
which will lead to a stop-the-world pause.
• Apache Cassandra uses a slab allocator internally.

54

Cassandra slab allocator

• 2MB slab sizes
• copy byte[] into them using compare-and-set
• GC before: 30-60 seconds every hour
• GC after: 5 seconds once in 3 days and 10 hours

55

Slab allocator constraints
• Works for limited usage:
• Buffers are written to linearly, flushed to disk and
recycled when they fill up.
• The objects need to be converted to binary
representation anyway.
• If you need random freeing and compaction, you’re
heading down the wrong direction.
• If you find yourself writing a full memory manager
on top of byte buffers, stop!
56

Soft references revisited

• Soft reference clearing is based on the amount of
free memory available when GC encounters the
reference.
• By deﬁnition, throughput collectors always clear
them.
• Can use them with CMS, but they increase memory
pressure and make the behavior less predictable.
• Need two GC cycles to get rid of referenced objects.
57

Everything More
than I ever wanted
to learned about
JVM performance
tuning
@twitter
Questions?
58

Attila Szegedi, Software Engineer
@asz

59

Everything I Ever Learned About JVM Performance Tuning @Twitter

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Viewers also liked

Viewers also liked (20)

Similar to Everything I Ever Learned About JVM Performance Tuning @Twitter

Similar to Everything I Ever Learned About JVM Performance Tuning @Twitter (20)

Recently uploaded

Recently uploaded (20)

Everything I Ever Learned About JVM Performance Tuning @Twitter

Editor's Notes