xref: /dpdk/doc/guides/prog_guide/mempool_lib.rst (revision 885807ae0785351c5a7a55d7ff44f7693f40b08d)
1..  SPDX-License-Identifier: BSD-3-Clause
2    Copyright(c) 2010-2014 Intel Corporation.
3
4.. _Mempool_Library:
5
6Mempool Library
7===============
8
9A memory pool is an allocator of a fixed-sized object.
10In the DPDK, it is identified by name and uses a mempool handler to store free objects.
11The default mempool handler is ring based.
12It provides some other optional services such as a per-core object cache and
13an alignment helper to ensure that objects are padded to spread them equally on all DRAM or DDR3 channels.
14
15This library is used by the :ref:`Mbuf Library <Mbuf_Library>`.
16
17Cookies
18-------
19
20In debug mode, cookies are added at the beginning and end of allocated blocks.
21The allocated objects then contain overwrite protection fields to help debugging buffer overflows.
22
23Debug mode is disabled by default,
24but can be enabled by setting ``RTE_LIBRTE_MEMPOOL_DEBUG`` in ``config/rte_config.h``.
25
26Stats
27-----
28
29In stats mode, statistics about get from/put in the pool are stored in the mempool structure.
30Statistics are per-lcore to avoid concurrent access to statistics counters.
31
32Stats mode is disabled by default,
33but can be enabled by setting ``RTE_LIBRTE_MEMPOOL_STATS`` in ``config/rte_config.h``.
34
35Memory Alignment Constraints on x86 architecture
36------------------------------------------------
37
38Depending on hardware memory configuration on X86 architecture, performance can be greatly improved by adding a specific padding between objects.
39The objective is to ensure that the beginning of each object starts on a different channel and rank in memory so that all channels are equally loaded.
40
41This is particularly true for packet buffers when doing L3 forwarding or flow classification.
42Only the first 64 bytes are accessed, so performance can be increased by spreading the start addresses of objects among the different channels.
43
44The number of ranks on any DIMM is the number of independent sets of DRAMs that can be accessed for the full data bit-width of the DIMM.
45The ranks cannot be accessed simultaneously since they share the same data path.
46The physical layout of the DRAM chips on the DIMM itself does not necessarily relate to the number of ranks.
47
48When running an application, the EAL command line options provide the ability to add the number of memory channels and ranks.
49
50.. note::
51
52    The command line must always have the number of memory channels specified for the processor.
53
54Examples of alignment for different DIMM architectures are shown in
55:numref:`figure_memory-management` and :numref:`figure_memory-management2`.
56
57.. _figure_memory-management:
58
59.. figure:: img/memory-management.*
60
61   Two Channels and Quad-ranked DIMM Example
62
63
64In this case, the assumption is that a packet is 16 blocks of 64 bytes, which is not true.
65
66The Intel® 5520 chipset has three channels, so in most cases,
67no padding is required between objects (except for objects whose size are n x 3 x 64 bytes blocks).
68
69.. _figure_memory-management2:
70
71.. figure:: img/memory-management2.*
72
73   Three Channels and Two Dual-ranked DIMM Example
74
75
76When creating a new pool, the user can specify to use this feature or not.
77
78.. note::
79
80   This feature is not present for Arm systems.
81   Modern Arm Interconnects choose the SN-F (memory channel)
82   using a hash of memory address bits.
83   As a result, the load is distributed evenly in all cases,
84   including the above described, rendering this feature unnecessary.
85
86
87.. _mempool_local_cache:
88
89Local Cache
90-----------
91
92In terms of CPU usage, the cost of multiple cores accessing a memory pool's ring of free buffers may be high
93since each access requires a compare-and-set (CAS) operation.
94To avoid having too many access requests to the memory pool's ring,
95the memory pool allocator can maintain a per-core cache and do bulk requests to the memory pool's ring,
96via the cache with many fewer locks on the actual memory pool structure.
97In this way, each core has full access to its own cache (with locks) of free objects and
98only when the cache fills does the core need to shuffle some of the free objects back to the pools ring or
99obtain more objects when the cache is empty.
100
101While this may mean a number of buffers may sit idle on some core's cache,
102the speed at which a core can access its own cache for a specific memory pool without locks provides performance gains.
103
104The cache is composed of a small, per-core table of pointers and its length (used as a stack).
105This internal cache can be enabled or disabled at creation of the pool.
106
107The maximum size of the cache is static and is defined at compilation time (RTE_MEMPOOL_CACHE_MAX_SIZE).
108
109:numref:`figure_mempool` shows a cache in operation.
110
111.. _figure_mempool:
112
113.. figure:: img/mempool.*
114
115   A mempool in Memory with its Associated Ring
116
117Alternatively to the internal default per-lcore local cache, an application can create and manage external caches through the ``rte_mempool_cache_create()``, ``rte_mempool_cache_free()`` and ``rte_mempool_cache_flush()`` calls.
118These user-owned caches can be explicitly passed to ``rte_mempool_generic_put()`` and ``rte_mempool_generic_get()``.
119The ``rte_mempool_default_cache()`` call returns the default internal cache if any.
120In contrast to the default caches, user-owned caches can be used by unregistered non-EAL threads too.
121
122.. _Mempool_Handlers:
123
124Mempool Handlers
125------------------------
126
127This allows external memory subsystems, such as external hardware memory
128management systems and software based memory allocators, to be used with DPDK.
129
130There are two aspects to a mempool handler.
131
132* Adding the code for your new mempool operations (ops). This is achieved by
133  adding a new mempool ops code, and using the ``RTE_MEMPOOL_REGISTER_OPS`` macro.
134
135* Using the new API to call ``rte_mempool_create_empty()`` and
136  ``rte_mempool_set_ops_byname()`` to create a new mempool and specifying which
137  ops to use.
138
139Several different mempool handlers may be used in the same application. A new
140mempool can be created by using the ``rte_mempool_create_empty()`` function,
141then using ``rte_mempool_set_ops_byname()`` to point the mempool to the
142relevant mempool handler callback (ops) structure.
143
144Legacy applications may continue to use the old ``rte_mempool_create()`` API
145call, which uses a ring based mempool handler by default. These applications
146will need to be modified to use a new mempool handler.
147
148For applications that use ``rte_pktmbuf_create()``, there is a config setting
149(``RTE_MBUF_DEFAULT_MEMPOOL_OPS``) that allows the application to make use of
150an alternative mempool handler.
151
152  .. note::
153
154    When running a DPDK application with shared libraries, mempool handler
155    shared objects specified with the '-d' EAL command-line parameter are
156    dynamically loaded. When running a multi-process application with shared
157    libraries, the -d arguments for mempool handlers *must be specified in the
158    same order for all processes* to ensure correct operation.
159
160
161Use Cases
162---------
163
164All allocations that require a high level of performance should use a pool-based memory allocator.
165Below are some examples:
166
167*   :ref:`Mbuf Library <Mbuf_Library>`
168
169*   :ref:`Environment Abstraction Layer <Environment_Abstraction_Layer>` , for logging service
170
171*   Any application that needs to allocate fixed-sized objects in the data plane and that will be continuously utilized by the system.
172