1 ==================================================
2 memaslap - Load testing and benchmarking a server
3 ==================================================
21 memaslap is a load generation and benchmark tool for :program:`memcached()`
22 servers. It generates configurable workload such as threads, concurrencies, connections, run time, overwrite, miss rate, key size, value size, get/set proportion, expected throughput, and so on. Furthermore, it also testss data
23 verification, expire-time verification, UDP, binary protocol, facebook test,
24 replication test, multi-get and reconnection, etc.
26 Memslap manages network connections like memcached with
27 libevent. Each thread of memaslap is bound with a CPU core, all
28 the threads don't communicate with each other, and there are several socket
29 connections in each thread. Each connection keeps key size distribution,
30 value size distribution, and command distribution by itself.
32 You can specify servers via the :option:`--servers` option or via the
33 environment variable :envvar:`MEMCACHED_SERVERS`.
41 Memslap is developed to for the following purposes:
44 Manages network connections with libevent asynchronously.
48 Set both TCP and UDP up to use non-blocking IO.
52 Improves parallelism: higher performance in multi-threads environments.
56 Improves time efficiency: faster processing speed.
60 Generates key and value more efficiently; key size distribution and value size distribution are configurable.
64 Supports get, multi-get, and set commands; command distribution is configurable.
68 Supports controllable miss rate and overwrite rate.
72 Supports data and expire-time verification.
76 Supports dumping statistic information periodically.
80 Supports thousands of TCP connections.
84 Supports binary protocol.
88 Supports facebook test (set with TCP and multi-get with UDP) and replication test.
98 Effective implementation of network.
99 ____________________________________
102 For memaslap, both TCP and UDP use non-blocking network IO. All
103 the network events are managed by libevent as memcached. The network module
104 of memaslap is similar to memcached. Libevent can ensure
105 memaslap can handle network very efficiently.
108 Effective implementation of multi-threads and concurrency
109 _________________________________________________________
112 Memslap has the similar implementation of multi-threads to
113 memcached. Memslap creates one or more self-governed threads;
114 each thread is bound with one CPU core if the system testss setting CPU
117 In addition, each thread has a libevent to manage the events of the network;
118 each thread has one or more self-governed concurrencies; and each
119 concurrency has one or more socket connections. All the concurrencies don’t
120 communicate with each other even though they are in the same thread.
122 Memslap can create thousands of socket connections, and each
123 concurrency has tens of socket connections. Each concurrency randomly or
124 sequentially selects one socket connection from its socket connection pool
125 to run, so memaslap can ensure each concurrency handles one
126 socket connection at any given time. Users can specify the number of
127 concurrency and socket connections of each concurrency according to their
131 Effective implementation of generating key and value
132 ____________________________________________________
135 In order to improve time efficiency and space efficiency,
136 memaslap creates a random characters table with 10M characters. All the
137 suffixes of keys and values are generated from this random characters table.
139 Memslap uses the offset in the character table and the length
140 of the string to identify a string. It can save much memory.
141 Each key contains two parts, a prefix and a suffix. The prefix is an
142 uint64_t, 8 bytes. In order to verify the data set before,
143 memaslap need to ensure each key is unique, so it uses the prefix to identify
144 a key. The prefix cannot include illegal characters, such as ‘\r’, ‘\n’,
145 ‘\0’ and ‘ ‘. And memaslap has an algorithm to ensure that.
147 Memslap doesn’t generate all the objects (key-value pairs) at
148 the beginning. It only generates enough objects to fill the task window
149 (default 10K objects) of each concurrency. Each object has the following
150 basic information, key prefix, key suffix offset in the character table, key
151 length, value offset in the character table, and value length.
153 In the work process, each concurrency sequentially or randomly selects an
154 object from the window to do set operation or get operation. At the same
155 time, each concurrency kicks objects out of its window and adds new object
159 Simple but useful task scheduling
160 _________________________________
163 Memslap uses libevent to schedule all the concurrencies of
164 threads, and each concurrency schedules tasks based on the local task
165 window. Memslap assumes that if each concurrency keeps the same
166 key distribution, value distribution and commands distribution, from
167 outside, memaslap keeps all the distribution as a whole.
168 Each task window includes a lot of objects, each object stores its basic
169 information, such as key, value, expire time, and so on. At any time, all
170 the objects in the window keep the same and fixed key and value
171 distribution. If an object is overwritten, the value of the object will be
172 updated. Memslap verifies the data or expire-time according to
173 the object information stored in the task window.
175 Libevent selects which concurrency to handle based on a specific network
176 event. Then the concurrency selects which command (get or set) to operate
177 based on the command distribution. If it needs to kick out an old object and
178 add a new object, in order to keep the same key and value distribution, the
179 new object must have the same key length and value length.
181 If memcached server has two cache layers (memory and SSD), running
182 memaslap with different window sizes can get different cache
183 miss rates. If memaslap adds enough objects into the windows at
184 the beginning, and the cache of memcached cannot store all the objects
185 initialized, then memaslap will get some objects from the second
186 cache layer. It causes the first cache layer to miss. So the user can
187 specify the window size to get the expected miss rate of the first cache
191 Useful implementation of multi-servers , UDP, TCP, multi-get and binary protocol
192 ________________________________________________________________________________
195 Because each thread is self-governed, memaslap can assign
196 different threads to handle different memcached servers. This is just one of
197 the ways in which memaslap testss multiple servers. The only
198 limitation is that the number of servers cannot be greater than the number
199 of threads. The other way to tests multiple servers is for replication
200 test. Each concurrency has one socket connection to each memcached server.
201 For the implementation, memaslap can set some objects to one
202 memcached server, and get these objects from the other servers.
204 By default, Memslap does single get. If the user specifies
205 multi-get option, memaslap will collect enough get commands and
206 pack and send the commands together.
208 Memslap testss both the ASCII protocol and binary protocol,
209 but it runs on the ASCII protocol by default.
210 Memslap by default runs on the TCP protocol, but it also
211 testss UDP. Because UDP is unreliable, dropped packages and out-of-order
212 packages may occur. Memslap creates a memory buffer to handle
213 these problems. Memslap tries to read all the response data of
214 one command from the server and reorders the response data. If some packages
215 get lost, the waiting timeout mechanism can ensure half-baked packages will
216 be discarded and the next command will be sent.
225 Below are some usage samples:
228 memaslap -s 127.0.0.1:11211 -S 5s
232 memaslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
236 memaslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
240 memaslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
244 memaslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
248 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
252 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
256 The user must specify one server at least to run memaslap. The
257 rest of the parameters have default values, as shown below:
259 Thread number = 1 Concurrency = 16
261 Run time = 600 seconds Configuration file = NULL
263 Key size = 64 Value size = 1024
265 Get/set = 9:1 Window size = 10k
267 Execute number = 0 Single get = true
269 Multi-get = false Number of sockets of each concurrency = 1
271 Reconnect = false Data verification = false
273 Expire-time verification = false ASCII protocol = true
275 Binary protocol = false Dumping statistic information
279 Overwrite proportion = 0% UDP = false
281 TCP = true Limit throughput = false
283 Facebook test = false Replication test = false
285 Key size, value size and command distribution.
286 ______________________________________________
289 All the distributions are read from the configuration file specified by user
290 with “—cfg_cmd” option. If the user does not specify a configuration file,
291 memaslap will run with the default distribution (key size = 64,
292 value size = 1024, get/set = 9:1). For information on how to edit the
293 configuration file, refer to the “Configuration File” section.
295 The minimum key size is 16 bytes; the maximum key size is 250 bytes. The
296 precision of proportion is 0.001. The proportion of distribution will be
297 rounded to 3 decimal places.
299 The minimum value size is 1 bytes; the maximum value size is 1M bytes. The
300 precision of proportion is 0.001. The proportion of distribution will be
301 rounded to 3 decimal places.
302 Currently, memaslap only testss set and get commands. And it
303 testss 100% set and 100% get. For 100% get, it will preset some objects to
307 Multi-thread and concurrency
308 ____________________________
311 The high performance of memaslap benefits from the special
312 schedule of thread and concurrency. It’s important to specify the proper
313 number of them. The default number of threads is 1; the default number of
314 concurrency is 16. The user can use “—threads” and “--concurrency” to
315 specify these variables.
317 If the system testss setting CPU affinity and the number of threads
318 specified by the user is greater than 1, memaslap will try to
319 bind each thread to a different CPU core. So if you want to get the best
320 performance memaslap, it is better to specify the number of
321 thread equal to the number of CPU cores. The number of threads specified by
322 the user can also be less or greater than the number of CPU cores. Because
323 of the limitation of implementation, the number of concurrencies could be
324 the multiple of the number of threads.
326 1. For 8 CPU cores system
330 --threads=2 --concurrency=128
332 --threads=8 --concurrency=128
334 --threads=8 --concurrency=256
336 --threads=12 --concurrency=144
338 2. For 16 CPU cores system
342 --threads=8 --concurrency=128
344 --threads=16 --concurrency=256
346 --threads=16 --concurrency=512
348 --threads=24 --concurrency=288
350 The memaslap performs very well, when
351 used to test the performance of memcached servers.
352 Most of the time, the bottleneck is the network or
353 the server. If for some reason the user wants to
354 limit the performance of memaslap, there
355 are two ways to do this:
357 Decrease the number of threads and concurrencies.
358 Use the option “--tps” that memaslap
359 provides to limit the throughput. This option allows
360 the user to get the expected throughput. For
361 example, assume that the maximum throughput is 50
362 kops/s for a specific configuration, you can specify
363 the throughput equal to or less than the maximum
364 throughput using “--tps” option.
371 Most of the time, the user does not need to specify the window size. The
372 default window size is 10k. For Schooner Memcached, the user can specify
373 different window sizes to get different cache miss rates based on the test
374 case. Memslap testss cache miss rate between 0% and 100%.
375 If you use this utility to test the performance of Schooner Memcached, you
376 can specify a proper window size to get the expected cache miss rate. The
377 formula for calculating window size is as follows:
379 Assume that the key size is 128 bytes, and the value size is 2048 bytes, and
382 1. Small cache cache_size=1M, 100% cache miss (all data get from SSD).
387 (1). cache miss rate 0%
391 (2). cache miss rate 5%
397 (1). cache miss rate 0%
407 The formula for calculating window size for cache miss rate 0%:
409 cache_size / concurrency / (key_size + value_size) \* 0.5
411 The formula for calculating window size for cache miss rate 5%:
413 cache_size / concurrency / (key_size + value_size) \* 0.7
420 Memslap testss both data verification and expire-time
421 verification. The user can use "--verify=" or "-v" to specify the proportion
422 of data verification. In theory, it testss 100% data verification. The
423 user can use "--exp_verify=" or "-e" to specify the proportion of
424 expire-time verification. In theory, it testss 100% expire-time
425 verification. Specify the "--verbose" options to get more detailed error
428 For example: --exp_verify=0.01 –verify=0.1 , it means that 1% of the objects
429 set with expire-time, 10% of the objects gotten will be verified. If the
430 objects are gotten, memaslap will verify the expire-time and
434 multi-servers and multi-config
435 _______________________________
438 Memslap testss multi-servers based on self-governed thread.
439 There is a limitation that the number of servers cannot be greater than the
440 number of threads. Memslap assigns one thread to handle one
441 server at least. The user can use the "--servers=" or "-s" option to specify
446 --servers=10.1.1.1:11211,10.1.1.2:11212,10.1.1.3:11213 --threads=6 --concurrency=36
448 The above command means that there are 6 threads, with each thread having 6
449 concurrencies and that threads 0 and 3 handle server 0 (10.1.1.1); threads 1
450 and 4 handle server 1 (10.1.1.2); and thread 2 and 5 handle server 2
453 All the threads and concurrencies in memaslap are self-governed.
455 So is memaslap. The user can start up several
456 memaslap instances. The user can run memaslap on different client
457 machines to communicate with the same memcached server at the same. It is
458 recommended that the user start different memaslap on different
459 machines using the same configuration.
462 Run with execute number mode or time mode
463 _________________________________________
466 The default memaslap runs with time mode. The default run time
467 is 10 minutes. If it times out, memaslap will exit. Do not
468 specify both execute number mode and time mode at the same time; just
473 --time=30s (It means the test will run 30 seconds.)
475 --execute_number=100000 (It means that after running 100000 commands, the test will exit.)
478 Dump statistic information periodically.
479 ________________________________________
482 The user can use "--stat_freq=" or "-S" to specify the frequency.
488 Memslap will dump the statistics of the commands (get and set) at the frequency of every 20
491 For more information on the format of dumping statistic information, refer to “Format of Output” section.
498 The user can use "--division=" or "-d" to specify multi-get keys count.
499 Memslap by default does single get with TCP. Memslap also testss data
500 verification and expire-time verification for multi-get.
502 Memslap testss multi-get with both TCP and UDP. Because of
503 the different implementation of the ASCII protocol and binary protocol,
504 there are some differences between the two. For the ASCII protocol,
505 memaslap sends one “multi-get” to the server once. For the
506 binary protocol, memaslap sends several single get commands
507 together as “multi-get” to the server.
514 Memslap testss both UDP and TCP. For TCP,
515 memaslap does not reconnect the memcached server if socket connections are
516 lost. If all the socket connections are lost or memcached server crashes,
517 memaslap will exit. If the user specifies the “--reconnect”
518 option when socket connections are lost, it will reconnect them.
520 User can use “--udp” to enable the UDP feature, but UDP comes with some
523 UDP cannot set data more than 1400 bytes.
525 UDP is not testsed by the binary protocol because the binary protocol of
526 memcached does not tests that.
528 UDP doesn’t tests reconnection.
535 Set data with TCP and multi-get with UDP. Specify the following options:
537 "--facebook --division=50"
539 If you want to create thousands of TCP connections, specify the
541 "--conn_sock=" option.
543 For example: --facebook --division=50 --conn_sock=200
545 The above command means that memaslap will do facebook test,
546 each concurrency has 200 socket TCP connections and one UDP socket.
548 Memslap sets objects with the TCP socket, and multi-gets 50
549 objects once with the UDP socket.
551 If you specify "--division=50", the key size must be less that 25 bytes
552 because the UDP packet size is 1400 bytes.
559 For replication test, the user must specify at least two memcached servers.
560 The user can use “—rep_write=” option to enable feature.
564 --servers=10.1.1.1:11211,10.1.1.2:11212 –rep_write=2
566 The above command means that there are 2 replication memcached servers,
567 memaslap will set objects to both server 0 and server 1, get
568 objects which are set to server 0 before from server 1, and also get objects
569 which are set to server 1 before from server 0. If server 0 crashes,
570 memaslap will only get objects from server 1. If server 0 comes
571 back to life again, memaslap will reconnect server 0. If both
572 server 0 and server 1 crash, memaslap will exit.
575 Supports thousands of TCP connections
576 _____________________________________
579 Start memaslap with "--conn_sock=" or "-n" to enable this
580 feature. Make sure that your system can tests opening thousands of files
581 and creating thousands of sockets. However, this feature does not tests
582 reconnection if sockets disconnect.
586 --threads=8 --concurrency=128 --conn_sock=128
588 The above command means that memaslap starts up 8 threads, each
589 thread has 16 concurrencies, each concurrency has 128 TCP socket
590 connections, and the total number of TCP socket connections is 128 \* 128 =
594 Supports binary protocol
595 ________________________
598 Start memaslap with "--binary" or "-B" options to enable this
599 feature. It testss all the above features except UDP, because the latest
600 memcached 1.3.3 does not implement binary UDP protocol.
606 Since memcached 1.3.3 doesn't implement binary UDP protocol,
607 memaslap does not tests UDP. In addition, memcached 1.3.3 does not tests
608 multi-get. If you specify "--division=50" option, it just sends 50 get
609 commands together as “mulit-get” to the server.
618 This section describes the format of the configuration file. By default
619 when no configuration file is specified memaslap reads the default
620 one located at ~/.memaslap.cnf.
622 Below is a sample configuration file:
627 ---------------------------------------------------------------------------
628 #comments should start with '#'
630 #start_len end_len proportion
632 #key length range from start_len to end_len
633 #start_len must be equal to or greater than 16
634 #end_len must be equal to or less than 250
635 #start_len must be equal to or greater than end_len
636 #memaslap will generate keys according to the key range
637 #proportion: indicates keys generated from one range accounts for the total
640 #example1: key range 16~100 accounts for 80%
641 # key range 101~200 accounts for 10%
642 # key range 201~250 accounts for 10%
643 # total should be 1 (0.8+0.1+0.1 = 1)
649 #example2: all keys length are 128 bytes
655 #start_len end_len proportion
657 #value length range from start_len to end_len
658 #start_len must be equal to or greater than 1
659 #end_len must be equal to or less than 1M
660 #start_len must be equal to or greater than end_len
661 #memaslap will generate values according to the value range
662 #proportion: indicates values generated from one range accounts for the
663 total generated values
665 #example1: value range 1~1000 accounts for 80%
666 # value range 1001~10000 accounts for 10%
667 # value range 10001~100000 accounts for 10%
668 # total should be 1 (0.8+0.1+0.1 = 1)
674 #example2: all value length are 128 bytes
680 #cmd_type cmd_proportion
682 #currently memaslap only testss get and set command.
688 #example: set command accounts for 50%
689 # get command accounts for 50%
690 # total should be 1 (0.5+0.5 = 1)
706 At the beginning, memaslap displays some configuration information as follows:
709 servers : 127.0.0.1:11211
729 set proportion: set_prop=0.10
733 get proportion: get_prop=0.90
744 The servers used by memaslap.
750 The number of threads memaslap runs with.
756 The number of concurrencies memaslap runs with.
762 How long to run memaslap.
768 The task window size of each concurrency.
774 The proportion of set command.
780 The proportion of get command.
784 The output of dynamic statistics is something like this:
789 ---------------------------------------------------------------------------------------------------------------------------------
791 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
792 Avg(us) Std_dev Geo_dist
793 Period 5 345826 69165 65.3 0 27 2198 203
795 Global 20 1257935 62896 71.8 0 26 3791 224
800 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
801 Avg(us) Std_dev Geo_dist
802 Period 5 38425 7685 7.3 0 42 628 240
804 Global 20 139780 6989 8.0 0 37 3790 253
809 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
810 Avg(us) Std_dev Geo_dist
811 Period 5 384252 76850 72.5 0 27 2198 207
813 Global 20 1397720 69886 79.7 0 26 3791 227
815 ---------------------------------------------------------------------------------------------------------------------------------
826 Statistics information of get command
832 Statistics information of set command
838 Statistics information of both get and set command
844 Result within a period
862 Throughput, operations/second
874 How many objects can’t be gotten
880 The minimum response time
886 The maximum response time
892 The average response time
898 Standard deviation of response time
904 Geometric distribution based on natural exponential function
908 At the end, memaslap will output something like this:
913 ---------------------------------------------------------------------------------------------------------------------------------
914 Get Statistics (1257956 events)
922 8: 484890 459823 12543 824
925 Set Statistics (139782 events)
933 8: 50784 65574 2064 167
936 Total Statistics (1397738 events)
944 8: 535674 525397 14607 991
954 written_bytes: 242516030
955 read_bytes: 1003702556
956 object_bytes: 152086080
961 Run time: 20.0s Ops: 1397754 TPS: 69817 Net_rate: 59.4M/s
962 ---------------------------------------------------------------------------------------------------------------------------------
973 Get statistics of response time
979 Set statistics of response time
985 Both get and set statistics of response time
991 The accumulated and minimum response time
997 The accumulated and maximum response time
1003 The accumulated and average response time
1009 Standard deviation of response time
1015 Geometric distribution based on logarithm 2
1021 Total get commands done
1027 Total set commands done
1033 How many objects can’t be gotten from server
1039 How many objects need to verify but can’t get them
1045 How many objects with insistent value
1051 How many objects are expired but we get them
1057 How many objects are unexpired but we can’t get them
1081 How many UDP packages are disorder
1087 How many UDP packages are lost
1093 How many times UDP time out happen
1111 Throughput, operations/second
1117 The average rate of network
1129 List one or more servers to connect. Servers count must be less than
1130 threads count. e.g.: --servers=localhost:1234,localhost:11211
1133 Number of threads to startup, better equal to CPU numbers. Default 8.
1136 Number of concurrency to simulate with load. Default 128.
1139 Number of TCP socks per concurrency. Default 1.
1141 -x, --execute_number=
1142 Number of operations(get and set) to execute for the
1143 given test. Default 1000000.
1146 How long the test to run, suffix: s-seconds, m-minutes, h-hours,
1147 d-days e.g.: --time=2h.
1150 Load the configure file to get command,key and value distribution list.
1153 Task window size of each concurrency, suffix: K, M e.g.: --win_size=10k.
1157 Fixed length of value.
1160 The proportion of date verification, e.g.: --verify=0.01
1163 Number of keys to multi-get once. Default 1, means single get.
1166 Frequency of dumping statistic information. suffix: s-seconds,
1167 m-minutes, e.g.: --resp_freq=10s.
1170 The proportion of objects with expire time, e.g.: --exp_verify=0.01.
1171 Default no object with expire time
1174 The proportion of objects need overwrite, e.g.: --overwrite=0.01.
1175 Default never overwrite object.
1178 Reconnect tests, when connection is closed it will be reconnected.
1181 UDP tests, default memaslap uses TCP, TCP port and UDP port of
1182 server must be same.
1185 Whether it enables facebook test feature, set with TCP and multi-get with UDP.
1188 Whether it enables binary protocol. Default with ASCII protocol.
1191 Expected throughput, suffix: K, e.g.: --tps=10k.
1194 The first nth servers can write data, e.g.: --rep_write=2.
1197 Whether it outputs detailed information when verification fails.
1200 Display this message and then exit.
1203 Display the version of the application and then exit.
1211 memaslap -s 127.0.0.1:11211 -S 5s
1213 memaslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
1215 memaslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
1217 memaslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
1219 memaslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
1221 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
1223 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
1231 To find out more information please check:
1232 `http://libmemcached.org/ <http://libmemcached.org/>`_
1240 Mingqiang Zhuang <mingqiangzhuang@hengtiansoft.com> (Schooner Technolgy)
1241 Brian Aker, <brian@tangent.org>
1248 :manpage:`memcached(1)` :manpage:`libmemcached(3)`