The document discusses Memcached, a distributed memory caching system. It was originally developed to speed up LiveJournal but is now widely used by companies like Facebook, Twitter, Wikipedia and others to speed up dynamic database-driven websites. Memcached stores objects in memory to reduce database load and speed up access. It uses a simple API with basic operations like set, get, delete and supports high availability through replication across multiple servers.

1. About Memcached
E-mail lcycenter@gmail.com

2. Memcached?
Memcached LiveJournal
, ,
, Web
LiveJournal Memcached MogileFS Perlbal
: http://www.danga.com/memcached

3. ?

4. ?

5. Cache
Do one thing well

6. Map
key1: value1
key2:value2
key3:value3
Cache
Do one thing well

7. Map
key1: value1
key2:value2 /
key3:value3
Cache
Do one thing well

8. Map
key1: value1
key2:value2 / Expired
key3:value3
Cache
Do one thing well

9. Map
key1: value1
key2:value2 / Expired
key3:value3
Cache
Do one thing well
I am not...
persistent
failover / redundant
authentication
RDBMS...

10. Simple
• C/S
• support telnet interface
•
✴set / get / stats

11. Simple
• C/S
• support telnet interface
•
✴set / get / stats

12. Simple
• C/S
• support telnet interface
•
✴set / get / stats

13. Simple
• C/S
• support telnet interface
•
✴set / get / stats

15. Object getFoo(String fooId) {
Object foo = memcached.get("foo" + fooId);
if (foo != null) return foo;
foo = database.fetchFoo(fooId);
memcached.set("foo"+fooId, foo);
return foo;
}

16. Base Patten

17. any problem?
Base Patten

18. Prefetch Patten

19. any problem?
Prefetch Patten

20. if (memcached.get(key) == null) {
if (memcached.add(key_mutex, 3*60*1000) == true) {
foo = database.fetchFoo(key);
memcached.set(key, foo);
memcached.delete(key_mutex);
} else {
sleep(50);
retry();
}
}
Mutex Patten

21. Mutex Patten

22. any problem?
Mutex Patten

23. foo = memcached.get(key);
if (foo == null) {
if (memcached.add(key_mutex, 3 * 60 * 1000) == true) {
foo = database.get(key);
memcached.set(key, foo);
memcached.delete(key_mutex);
} else {
sleep(50);
retry();
}
} else {
...
}
Mutex Patten II

24. foo = memcached.get(key);
if (foo == null) {
if (memcached.add(key_mutex, 3 * 60 * 1000) == true) {
foo = database.get(key);
memcached.set(key, foo);
memcached.delete(key_mutex);
} else {
sleep(50);
retry();
}
} else {
... here
}
Mutex Patten II

25. if (foo.timeout >= now()) {
if (memcached.add(key_mutex, 3 * 60 * 1000) == true) {
foo.timeout += 3 * 60 * 1000;
memcached.set(key, foo, KEY_TIMEOUT * 2);
foo = database.get(key);
foo.timeout = KEY_TIMEOUT;
memcached.set(key, foo, KEY_TIMEOUT * 2);
memcached.delete(key_mutex);
} else {
sleep(50);
retry();
}
} else {
return foo;
}
Mutex Patten II

26. if (foo.timeout >= now()) {
if (memcached.add(key_mutex, 3 * 60 * 1000) == true) {
foo.timeout += 3 * 60 * 1000;
memcached.set(key, foo, KEY_TIMEOUT * 2);
foo = database.get(key);
foo.timeout = KEY_TIMEOUT;
memcached.set(key, foo, KEY_TIMEOUT * 2);
memcached.delete(key_mutex);
} else {
sleep(50);
retry();
}
} else {
return foo;
}
Mutex Patten II

27. Mutex Patten II

28. any problem?
Mutex Patten II

31. ?
WEB EJB RDBMS
HTML Marshall
SQL
JSON Java
Results
XML Objects

32. ?
WEB EJB RDBMS
HTML Marshall
SQL
JSON Java
Results
XML Objects

33. ?
WEB EJB RDBMS
HTML Marshall
SQL
JSON Java
Results
XML Objects

34. • (memcached exptimes)
•
- SET or DELETE
-
• version
- version key: user_v2_100
- + LRU

35. Distribute Memcached
• MC HASH
• MC Fail-over
• MC-MC
• DB-MC fail
• Key-Value DB

36. MC Hash
•
- MC
• Hash
-
-
-
•

37. idx = Zlib.crc32(key) % servers.size
Consistent Hashing

38. Consistent Hashing

39. Consistent Hashing

40. Others
•
• hot keys
• create local cache
• alias key
•

44. Memcached Client
• Marshall
•
• Multiget support
• Compressed support
• Async/Sync Modes
• Replicated
• ...

45. Stats
• Memcached:
• telnet
• : stats

46. Stats

47. Limits
• Key Size (250 bytes)
• Data Size (under 1 megabyte)
• 32bit/64bit (maximum size of the process)

48. Protocol Commands
• set / get / gets / replace / add / delete
• append / prepend
• increment / decrement
• cas (compare and swap atomic!)
• stats(detail)

49. Numbers Everyone
Should Know
L1 cache reference 0.5 ns
Branch mispredict 5 ns
L2 cache reference 7 ns
Mutex lock/unlock 100 ns
Main memory reference * 100 ns
Compress 1K bytes with Zippy 10,000 ns
Send 2K bytes over 1 Gbps network * 20,000 ns
Read 1 MB sequentially from memory * 250,000 ns
Round trip within same datacenter 500,000 ns
Disk seek * 10,000,000 ns
Read 1 MB sequentially from network * 10,000,000 ns
Read 1 MB sequentially from disk * 30,000,000 ns
Send packet CA->Netherlands->CA 150,000,000 ns