Workload Analysis of a LargeScale KeyValue Store Berk

Workload Analysis of a Large-Scale Key-Value Store Berk Atikoglu, Yuehai Xu, Eitan Fracthenberg, Song Yiang, Mike Paleczny

�+284, 000, 000 requests � 5 different use cases �Workload characteristics, locality, cache effectiveness 1

Database Web Servers Cache Servers 2

�Understand workload characteristics �Identify factors affecting performance �Provide a benchmark for future studies 3

�Distributed memory caching system �Key-value store for small objects Key Hash Function Memcached Servers 4

�Capture traces through a Linux Kernel Module (LKM) LKM Memcached Transport (TCP/UDP) Network Ethernet �Process traces with Hive 5

Contains server related information Pool Size Description USR Few User-account status information APP Dozens Object metadata of a popular application SYS Few System data on service location VAR Dozens Server-side browser information ETC Hundreds Nonspecific, general purpose Anything that doesn’t belong to a specific pool goes to ETC 6

�Workload Characteristics �Locality, Cache Behavior 7

> 99. 8% GET: UPDATE = 30: 1 8

90% of VAR keys are 31 B USR keys are 16 B or 21 B ETC is heterogeneous 9

USR values are only 2 B 90% of values are smaller than 500 B 10

90% of data is generated by values of 500 B or smaller except ETC 90% is 10 KB or smaller values for ETC 11

All pools show diurnal pattern except SYS 12

North America starts its day Night time in Western Semiphere 13

�Workload Characteristics �Locality, Cache Behavior 14

0. 0003% of keys in 10% of requests in ETC 1% of keys in 55% of requests in ETC Least frequent 50% of keys in 1% of requests in ETC 15

% of unique keys out of total in unit time 100 80 60 40 20 0 USR APP ETC 5 min VAR SYS 60 min 16

99. 9% of SYS keys 88. 5% of ETC keys are reused in 1 hr 96. 4% of ETC keys are reused in 6 hr 17

98. 2% 92. 9% 93. 7% Why? 81. 4% 98. 7% 18

hit Compulsory Capacity Invalidation 70% 8% 22% miss: compulsory miss: capacity miss: invalidation 4% 81% 19% 13% 2% 19

� Analyzed 5 different memcached use cases � Different applications of memcached have extreme variations in access patterns � Answered pertinent questions to improve Facebook’s memcached usage 20

�Questions? 21