Benchmarking, benchmarking, benchmarking. We all do it, mostly it tells us what we want to hear but often hides a mountain of misinformation. In this talk we will walk through the pitfalls that you might find yourself in by looking at some examples where things go wrong. We will then walk through how MongoDB performance is measured, the processes and methodology and ways to present and look at the information.

1. #MongoDBDays
Mythbusting: Understanding
How We Measure the
Performance of MongoDB
Alvin Richards
Senior Director of Performance Engineering, MongoDB

2. Before we start…
• We are going to look a lot at
– C++ kernel code
– Java benchmarks
– JavaScript tests
• And lots of charts
• And its going to be awesome!

3. Measuring "Performance"
https://www.youtube.com/watch?v=7wm-pZp_mi0

4. Benchmarking
• Some common traps
• Performance measurement & diagnosis
• What's next

5. Part One
Some Common Traps

6. "We all live in a house on fire, no fire department
to call; no way out, just the upstairs window to
look out of while the fire burns the house down
with us trapped, locked in it."
The Milk Train Doesn't Stop Here Anymore
Tennessee Williams

7. #1 Time taken to Insert x
Documents
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();

8. #1 Time taken to Insert x
Documents
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();

9. #1 Time taken to Insert x
Documents
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();

10. #1 Time taken to Insert x
Documents
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();

11. #1 Time taken to Insert x
Documents
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();

12. So that looks ok, right?
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();

13. What are else you measuring?
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();
Object creation and GC
management?

14. What are else you measuring?
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();
Object creation and GC
management?
Thread contention on
nextInt()?

15. What are else you measuring?
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();
Object creation and GC
management?
Thread contention on
nextInt()?
Time to synthesize data?

16. What are else you measuring?
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();
Object creation and GC
management?
Thread contention on
nextInt()?
Time to synthesize data?
Thread contention on
addAndGet()?

17. What are else you measuring?
long startTime = System.currentTimeMillis();
for (int roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
BasicDBObject doc = new BasicDBObject();
doc.put("_id",id);
doc.put("k",rand.nextInt(numMaxInserts)+1);
String cVal = "…"
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i]=doc;
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
globalInserts.addAndGet(documentsPerInsert);
}
long endTime = System.currentTimeMillis();
Object creation and GC
management?
Thread contention on
nextInt()?
Time to synthesize data?
Thread contention on
addAndGet()?
Clock resolution?

18. Solution: Pre-Create the objects
// Pre Create the Object outside the Loop
BasicDBObject[] aDocs = new BasicDBObject[documentsPerInsert];
for (int i=0; i < documentsPerInsert; i++) {
BasicDBObject doc = new BasicDBObject();
String cVal = "…";
doc.put("c",cVal);
String padVal = "…";
doc.put("pad",padVal);
aDocs[i] = doc;
}
Pre-create non varying
data outside the timing
loop
Alternative
• Pre-create the data in a file; load from file

19. Solution: Remove contention
// Use ThreadLocalRandom generator or an instance of java.util.Random per thread
java.util.concurrent.ThreadLocalRandom rand;
for (long roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
doc = aDocs[i];
doc.put("_id",id);
doc.put("k", nextInt(rand, numMaxInserts)+1);
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
}
// Maintain count outside the loop
globalInserts.addAndGet(documentsPerInsert * roundNum);
Remove contention
nextInt() by making
Thread local

20. Solution: Remove contention
// Use ThreadLocalRandom generator or an instance of java.util.Random per thread
java.util.concurrent.ThreadLocalRandom rand;
Remove contention
nextInt() by making
Thread local
for (long roundNum = 0; roundNum < numRounds; roundNum++) {
for (int i = 0; i < documentsPerInsert; i++) {
id++;
doc = aDocs[i];
doc.put("_id",id);
doc.put("k", nextInt(rand, numMaxInserts)+1);
}
coll.insert(aDocs);
numInserts += documentsPerInsert;
}
// Maintain count outside the loop
globalInserts.addAndGet(documentsPerInsert * roundNum);
Remove contention on
addAndGet()

21. Solution: Timer resolution
long startTime = System.currentTimeMillis();
…
long endTime = System.currentTimeMillis();
long startTime = System.nanoTime();
…
long endTime = System.nanoTime() - startTime;
"granularity of the value
depends on the
underlying operating
system and may be
larger"
"resolution is at least as
good as that of
currentTimeMillis()"
Source
• http://docs.oracle.com/javase/7/docs/api/java/lang/System.html

22. General Principal #1
Know what you are
measuring

23. #2 Response time to return all
results
BasicDBObject doc = new BasicDBObject();
doc.put("v", str); // str is a 2k string
for (int i=0; i < 1000; i++) {
doc.put("_id",i); coll.insert(doc);
}
BasicDBObject predicate = new BasicDBObject();
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate);
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();

24. #2 Response time to return all
results
BasicDBObject doc = new BasicDBObject();
doc.put("v", str); // str is a 2k string
for (int i=0; i < 1000; i++) {
doc.put("_id",i); coll.insert(doc);
}
BasicDBObject predicate = new BasicDBObject();
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate);
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();

25. #2 Response time to return all
results
BasicDBObject doc = new BasicDBObject();
doc.put("v", str); // str is a 2k string
for (int i=0; i < 1000; i++) {
doc.put("_id",i); coll.insert(doc);
}
BasicDBObject predicate = new BasicDBObject();
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate);
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();

26. #2 Response time to return all
results
BasicDBObject doc = new BasicDBObject();
doc.put("v", str); // str is a 2k string
for (int i=0; i < 1000; i++) {
doc.put("_id",i); coll.insert(doc);
}
BasicDBObject predicate = new BasicDBObject();
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate);
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();

27. So that looks ok, right?
BasicDBObject doc = new BasicDBObject();
doc.put("v", str); // str is a 2k string
for (int i=0; i < 1000; i++) {
doc.put("_id",i); coll.insert(doc);
}
BasicDBObject predicate = new BasicDBObject();
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate);
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();

28. What are else you measuring?
BasicDBObject doc = new BasicDBObject();
doc.put("v", str); // str is a 2k string
for (int i=0; i < 1000; i++) {
doc.put("_id",i); coll.insert(doc);
}
BasicDBObject predicate = new BasicDBObject();
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate);
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();
Each doc is is 4080 bytes
on disk with powerOf2Sizes

29. What are else you measuring?
BasicDBObject doc = new BasicDBObject();
doc.put("v", str); // str is a 2k string
for (int i=0; i < 1000; i++) {
doc.put("_id",i); coll.insert(doc);
}
BasicDBObject predicate = new BasicDBObject();
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate);
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();
Each doc is is 4080 bytes
on disk with powerOf2Sizes
Unrestricted predicate?

30. What are else you measuring?
BasicDBObject doc = new BasicDBObject();
doc.put("v", str); // str is a 2k string
for (int i=0; i < 1000; i++) {
doc.put("_id",i); coll.insert(doc);
}
BasicDBObject predicate = new BasicDBObject();
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate);
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();
Each doc is is 4080 bytes
on disk with powerOf2Sizes
Unrestricted predicate?
Measuring
• Time to parse &
execute query
• Time to retrieve all
document
But also
• Cost of shipping ~4MB
data through network
stack

31. Solution: Limit the projection
BasicDBObject predicate = new BasicDBObject();
predicate.put("_id", new BasicDBObject("$gte", 10).append("$lte", 20));
BasicDBObject projection = new BasicDBObject();
projection.put("_id", 1);
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate, projection );
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();
Return fixed range

32. Solution: Limit the projection
BasicDBObject predicate = new BasicDBObject();
predicate.put("_id", new BasicDBObject("$gte", 10).append("$lte", 20));
BasicDBObject projection = new BasicDBObject();
projection.put("_id", 1);
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate, projection );
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();
Return fixed range
Only project _id

33. Solution: Limit the projection
BasicDBObject predicate = new BasicDBObject();
predicate.put("_id", new BasicDBObject("$gte", 10).append("$lte", 20));
BasicDBObject projection = new BasicDBObject();
projection.put("_id", 1);
long startTime = System.currentTimeMillis();
DBCursor cur = coll.find(predicate, projection );
DBObject foundObj;
while (cur.hasNext()) {
foundObj = cur.next();
}
long endTime = System.currentTimeMillis();
Return fixed range
Only project _id
Only 46k transferred
through network stack

34. General Principal #2
Measure only what you
need to measure

35. Part Two
Performance
measurement &
diagnosis

36. "Every experiment destroys some of the
knowledge of the system which was obtained by
previous experiments."
The Physical Principles of the Quantum Theory (1930)
Werner Heisenberg

37. Broad categories
• Micro Benchmarks
• Workloads

38. Micro benchmarks: mongo-perf

39. mongo-perf: goals
• Measure
– commands
• Configure
– Single mongod, ReplSet size (1 -> n), Sharding
– Single vs. Multiple DB
– O/S
• Characterize
– Throughput by thread count
• Compare

40. What do you get?
Better

41. What do you get?
Measured
improvement
between rc0 and
rc2
Better

42. Benchmark source code
tests.push( { name: "Commands.CountsIntIDRange",
pre: function( collection ) {
collection.drop();
for ( var i = 0; i < 1000; i++ ) {
collection.insert( { _id : i } );
}
collection.getDB().getLastError();
},
ops: [
{ op: "command",
ns : "testdb",
command : { count : "mycollection",
query : { _id : { "$gt" : 10, "$lt" : 100 } } } }
] } );

43. Benchmark source code
tests.push( { name: "Commands.CountsIntIDRange",
pre: function( collection ) {
collection.drop();
for ( var i = 0; i < 1000; i++ ) {
collection.insert( { _id : i } );
}
collection.getDB().getLastError();
},
ops: [
{ op: "command",
ns : "testdb",
command : { count : "mycollection",
query : { _id : { "$gt" : 10, "$lt" : 100 } } } }
] } );

44. Benchmark source code
tests.push( { name: "Commands.CountsIntIDRange",
pre: function( collection ) {
collection.drop();
for ( var i = 0; i < 1000; i++ ) {
collection.insert( { _id : i } );
}
collection.getDB().getLastError();
},
ops: [
{ op: "command",
ns : "testdb",
command : { count : "mycollection",
query : { _id : { "$gt" : 10, "$lt" : 100 } } } }
] } );

45. Benchmark source code
tests.push( { name: "Commands.CountsIntIDRange",
pre: function( collection ) {
collection.drop();
for ( var i = 0; i < 1000; i++ ) {
collection.insert( { _id : i } );
}
collection.getDB().getLastError();
},
ops: [
{ op: "command",
ns : "testdb",
command : { count : "mycollection",
query : { _id : { "$gt" : 10, "$lt" : 100 } } } }
] } );

46. Code Change

47. Workloads
• "public" workloads
– YCSB
– Sysbench
• "real world" simulations
– Inbox fan in/out
– Message Stores
– Content Management

48. Example: Bulk Load Performance
16m Documents
Better
55% degradation
2.6.0-rc1 vs 2.4.10

49. Ouch… where's the tree in the
woods?
• 2.4.10 -> 2.6.0
– 4495 git commits

50. git-bisect
• Bisect between good/bad hashes
• git-bisect nominates a new githash
– Build against githash
– Re-run test
– Confirm if this githash is good/bad
• Rinse and repeat

51. Code Change - Bad Githash

52. Code Change - Fix

53. Bulk Load Performance - Fix
Better
11% improvement
2.6.1 vs 2.4.10

54. The problem with measurement
• Observability
– What can you observe on the system?
• Effect
– What effects does the observation cause?

55. mtools

56. mtools
• MongoDB log file analysis
– Filter logs for operations, events
– Response time, lock durations
– Plot
• https://github.com/rueckstiess/mtools

57. Response Times > 100ms
Bulk Insert 2.6.0-rc0
Ops/Sec
Time

58. Response Times > 100ms
Bulk Insert 2.6.0-rc0 vs. 2.6.0-rc2
Floor raised

59. Code Change – Yielding Policy

60. Code Change

61. Response Times
Bulk Insert 2.6.0 vs 2.6.1
Ceiling similar, lower floor
resulting in 40%
improvement in throughput

62. Secondary effects of Yield policy change
Write lock time reduced
Order of magnitude reduction
of write lock duration

63. Unexpected side effects of
measurement?
> db.serverStatus()
Yes – will cause a read lock to be acquired
> db.serverStatus({recordStats:0})
No – lock is not acquired
> mongostat
Yes - until SERVER-14008 resolved, uses db.serverStatus()

64. CPU sampling
• Get an impression of
– Call Graphs
– CPU time spent on node and called nodes

65. Setup & building with google-profiler
> sudo apt-get install google-perftools
> sudo apt-get install libunwind7-dev
> scons --use-cpu-profiler mongod

66. Start the profiling
> mongodb –dbpath <…>
Note: Do not use –fork
> mongo
> use admin
> db.runCommand({_cpuProfilerStart: {profileFilename: 'foo.prof'}})
Execute some commands that you want to profile
> db.runCommand({_cpuProfilerStop: 1})

67. Sample start vs. end of workload

68. Sample start vs. end of workload

69. Code change

70. Public Benchmarks – Not all forks are
the same…
• YCSB
– https://github.com/achille/YCSB
• sysbench-mongodb
– https://github.com/mdcallag/sysbench-mongodb

71. Part Three
And next?

72. "The future sucks. Change it."
"I'm way cool Beavis, but I cannot change the
future."
Beavis & Butthead

73. What we are working on
• mongo-perf
– UI refactor
– Adding more micro benchmarks (geo, sharding)
• Workloads
– Adding external benchmarks
– Creating benchmarks for common use cases
• Inbox fan in/out
• Analytical dashboards
• Stream / Feeds
• Customers, Partners & Community

74. Here's how you can help change the
future!
• Got a great workload? Great benchmark?
• Want to donate it?
• alvin@mongodb.com

75. Don't be that benchmark…
#1 Know what you are measuring
#2 Measure only what you need to
measure

76. #MongoDBDays
Thank You
Alvin Richards
alvin@mongodb.com / @jonnyeight
Senior Director of Performance Engineering, MongoDB