Changes in / [bb7c77d:2e9b59b]


Ignore:
Location:
doc/theses/mubeen_zulfiqar_MMath
Files:
203 added
4 edited

Legend:

Unmodified
Added
Removed

  • doc/theses/mubeen_zulfiqar_MMath/allocator.tex

    rbb7c77d r2e9b59b  
    2424%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    2525
     26<<<<<<< HEAD
     27\section{Design choices for uHeap}\label{sec:allocatorSec}
     28uHeap's design was reviewed and changed to fulfill new requirements (FIX ME: cite allocator philosophy). For this purpose, following two designs of uHeapLmm were proposed:
     29=======
    2630\section{Design Choices}
     31>>>>>>> bb7c77dc425e289ed60aa638529b3e5c7c3e4961
    2732
    2833llheap's design was reviewed and changed multiple times throughout the thesis.
    … …  
    248253\begin{figure}
    249254\centering
     255<<<<<<< HEAD
     256\includegraphics[width=0.65\textwidth]{figures/NewHeapStructure.eps}
     257\caption{uHeap Structure}
     258\label{fig:heapStructureFig}
     259=======
    250260% \includegraphics[width=0.65\textwidth]{figures/NewHeapStructure.eps}
    251261\input{llheap}
    252262\caption{llheap Structure}
    253263\label{f:llheapStructure}
     264>>>>>>> bb7c77dc425e289ed60aa638529b3e5c7c3e4961
    254265\end{figure}
    255266
    … …  
    331342\end{algorithm}
    332343
     344<<<<<<< HEAD
     345Algorithm~\ref{alg:heapObjectFreeOwn} shows how a free request is fulfilled if object ownership is turned on. Algorithm~\ref{alg:heapObjectFreeNoOwn} shows how the same free request is fulfilled without object ownership.
     346
     347\begin{algorithm}
     348\caption{Dynamic object free at address A with object ownership}\label{alg:heapObjectFreeOwn}
     349\begin{algorithmic}[1]
     350\If {$\textit{A was mmap-ed}$}
     351        \State $\text{return A's dynamic memory to system using system call munmap}$
     352=======
    333353\vspace*{-15pt}
    334354\begin{algorithm}[H]
    … …  
    338358\If {$\textit{A mapped allocation}$}
    339359        \State $\text{return A's dynamic memory to system using system call \lstinline{munmap}}$
     360>>>>>>> bb7c77dc425e289ed60aa638529b3e5c7c3e4961
    340361\Else
    341362        \State $\text{B} \gets \textit{O's owner}$
    … …  
    348369\end{algorithmic}
    349370\end{algorithm}
     371<<<<<<< HEAD
     372
     373\begin{algorithm}
     374\caption{Dynamic object free at address A without object ownership}\label{alg:heapObjectFreeNoOwn}
     375\begin{algorithmic}[1]
     376\If {$\textit{A was mmap-ed}$}
     377        \State $\text{return A's dynamic memory to system using system call munmap}$
     378\Else
     379        \State $\text{B} \gets \textit{O's owner}$
     380        \If {$\textit{B is thread-local heap's bucket}$}
     381                \State $\text{push A to B's free-list}$
     382        \Else
     383                \State $\text{C} \gets \textit{thread local heap's bucket with same size as B}$
     384                \State $\text{push A to C's free-list}$
     385        \EndIf
     386\EndIf
     387\end{algorithmic}
     388\end{algorithm}
     389
     390=======
     391>>>>>>> bb7c77dc425e289ed60aa638529b3e5c7c3e4961
    350392
    351393\vspace*{-15pt}

  • doc/theses/mubeen_zulfiqar_MMath/background.tex

    rbb7c77d r2e9b59b  
    757757Implementing lock-free operations for more complex data-structures (queue~\cite{Valois94}/deque~\cite{Sundell08}) is correspondingly more complex.
    758758Michael~\cite{Michael04} and Gidenstam \etal \cite{Gidenstam05} have created lock-free variations of the Hoard allocator.
     759
     760
     761\subsubsection{Speed Workload}
     762The worload method uses the opposite approach. It calls the allocator's routines for a specific amount of time and measures how much work was done during that time. Then, similar to the time method, it divides the time by the workload done during that time and calculates the average time taken by the allocator's routine.
     763*** FIX ME: Insert a figure of above benchmark with description
     764
     765\paragraph{Knobs}
     766*** FIX ME: Insert Knobs

  • doc/theses/mubeen_zulfiqar_MMath/benchmarks.tex

    rbb7c77d r2e9b59b  
    11\chapter{Benchmarks}
    22
    3 \noindent
    4 ====================
    5 
    6 Writing Points:
     3%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     4%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     5%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Micro Benchmark Suite
     6%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     7%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     8
     9The aim of micro benchmark suite is to create a set of programs that can evaluate a memory allocator based on the
     10performance matrices described in (FIX ME: local cite). These programs can be taken as a standard to benchmark an
     11allocator's basic goals. These programs give details of an allocator's memory overhead and speed under a certain
     12allocation pattern. The speed of the allocator is benchmarked in different ways. Similarly, false sharing happening in
     13an allocator is also measured in multiple ways. These benchmarks evalute the allocator under a certain allocation
     14pattern which is configurable and can be changed using a few knobs to benchmark observe an allocator's performance
     15under a desired allocation pattern.
     16
     17Micro Benchmark Suite benchmarks an allocator's performance by allocating dynamic objects and, then, measuring specifc
     18matrices. The benchmark suite evaluates an allocator with a certain allocation pattern. Bnechmarks have different knobs
     19that can be used to change allocation pattern and evaluate an allocator under desired conditions. These can be set by
     20giving commandline arguments to the benchmark on execution.
     21
     22\section{Current Benchmarks} There are multiple benchmarks that are built individually and evaluate different aspects of
     23 a memory allocator. But, there is not a set of benchamrks that can be used to evaluate multiple aspects of memory
     24 allocators.
     25
     26\subsection{threadtest}(FIX ME: cite benchmark and hoard) Each thread repeatedly allocates and then deallocates 100,000
     27 objects. Runtime of the benchmark evaluates its efficiency.
     28
     29\subsection{shbench}(FIX ME: cite benchmark and hoard) Each thread allocates and randomly frees a number of random-sized
     30 objects. It is a stress test that also uses runtime to determine efficiency of the allocator.
     31
     32\subsection{larson}(FIX ME: cite benchmark and hoard) Larson simulates a server environment. Multiple threads are
     33 created where each thread allocator and free a number of objects within a size range. Some objects are passed from
     34 threads to the child threads to free. It caluculates memory operations per second as an indicator of memory
     35 allocator's performance.
     36
     37\section{Memory Benchmark} Memory benchmark measures memory overhead of an allocator. It allocates a number of dynamic
     38 objects. Then, by reading /self/proc/maps, gets the total memory that the allocator has reuested from the OS. It
     39 calculates the memory head by taking the difference between the memory the allocator has requested from the OS and the
     40 memory that program has allocated.
     41
     42\begin{figure}
     43\centering
     44\includegraphics[width=1\textwidth]{figures/bench-memory.eps}
     45\caption{Benchmark Memory Overhead}
     46\label{fig:benchMemoryFig}
     47\end{figure}
     48
     49Figure \ref{fig:benchMemoryFig} gives a flow of the memory benchmark. It creates a producer-consumer scenerio with K producers
     50 and each producer has M consumers. Producer has a separate buffer for each consumer. Producer allocates N objects of
     51 random sizes following the given distrubution for each consumer. Consumer frees those objects. After every memory
     52 operation, program memory usage is recorded throughout the runtime. This data then can be used to visualize the memory
     53 usage and consumption of the prigram.
     54
     55Different knobs can be adjusted to set certain thread model.\\
     56-threadA :  sets number of alloc threads (producers) for mem benchmark\\
     57-consumeS:  sets production and conumption round size\\
     58-threadF :  sets number of free threads (consumers) for mem benchmark
     59
     60Object allocation size can be changed using the knobs:\\
     61-maxS    :  sets max object size\\
     62-minS    :  sets min object size\\
     63-stepS   :  sets object size increment\\
     64-distroS :  sets object size distribution\\
     65-objN    :  sets number of objects per thread\\
     66
     67\section{Speed Benchmark} Speed benchmark measures the runtime speed of an allocator (FIX ME: cite allocator routines).
     68 Speed benchmark measures runtime speed of individual memory allocation routines. It also considers different
     69 allocation chains to measures the performance of the allocator by combining multiple allocation routines in a chain.
     70 It uses following chains and measures allocator runtime speed against them:
    771\begin{itemize}
    8 \item
    9 Performance matrices of memory allocation.
    10 \item
    11 Aim of micro benchmark suite.
    12 
    13 ----- SHOULD WE GIVE IMPLEMENTATION DETAILS HERE? -----
    14 
    15 \PAB{For the benchmarks, yes.}
    16 \item
    17 A complete list of benchmarks in micro benchmark suite.
    18 \item
    19 One detailed section for each benchmark in micro benchmark suite including:
    20 
    21 \begin{itemize}
    22 \item
    23 The introduction of the benchmark.
    24 \item
    25 Figure.
    26 \item
    27 Results with popular memory allocators.
     72\item malloc 0
     73\item free NULL
     74\item malloc
     75\item realloc
     76\item free
     77\item calloc
     78\item malloc-free
     79\item realloc-free
     80\item calloc-free
     81\item malloc-realloc
     82\item calloc-realloc
     83\item malloc-realloc-free
     84\item calloc-realloc-free
     85\item malloc-realloc-free-calloc
    2886\end{itemize}
    2987
    30 \item
    31 Summarize performance of current memory allocators.
    32 \end{itemize}
    33 
    34 \noindent
    35 ====================
    36 
    37 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    38 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    39 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Performance Matrices
    40 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    41 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    42 
    43 
    44 \section{Benchmarks}
    45 There are multiple benchmarks that are built individually and evaluate different aspects of a memory allocator. But, there is not standard set of benchamrks that can be used to evaluate multiple aspects of memory allocators.
    46 
    47 \paragraph{threadtest}
    48 (FIX ME: cite benchmark and hoard) Each thread repeatedly allocates and then deallocates 100,000 objects. Runtime of the benchmark evaluates its efficiency.
    49 
    50 \paragraph{shbench}
    51 (FIX ME: cite benchmark and hoard) Each thread allocates and randomly frees a number of random-sized objects. It is a stress test that also uses runtime to determine efficiency of the allocator.
    52 
    53 \paragraph{larson}
    54 (FIX ME: cite benchmark and hoard) Larson simulates a server environment. Multiple threads are created where each thread allocator and free a number of objects within a size range. Some objects are passed from threads to the child threads to free. It caluculates memory operations per second as an indicator of memory allocator's performance.
    55 
    56 
    57 \section{Performance Matrices of Memory Allocators}
    58 
    59 When it comes to memory allocators, there are no set standards of performance. Performance of a memory allocator depends highly on the usage pattern of the application. A memory allocator that is the best performer for a certain application X might be the worst for some other application which has completely different memory usage pattern compared to the application X. It is extremely difficult to make one universally best memory allocator which will outperform every other memory allocator for every usage pattern. So, there is a lack of a set of standard benchmarks that are used to evaluate a memory allocators's performance.
    60 
    61 If we breakdown the goals of a memory allocator, there are two basic matrices on which a memory allocator's performance is evaluated.
    62 \begin{enumerate}
    63 \item
    64 Memory Overhead
    65 \item
    66 Speed
    67 \end{enumerate}
    68 
    69 \subsection{Memory Overhead}
    70 Memory overhead is the extra memory that a memory allocator takes from OS which is not requested by the application. Ideally, an allocator should get just enough memory from OS that can fulfill application's request and should return this memory to OS as soon as applications frees it. But, allocators retain more memory compared to what application has asked for which causes memory overhead. Memory overhead can happen for various reasons.
    71 
    72 \subsubsection{Fragmentation}
    73 Fragmentation is one of the major reasons behind memory overhead. Fragmentation happens because of situations that are either necassary for proper functioning of the allocator such as internal memory management and book-keeping or are out of allocator's control such as application's usage pattern.
    74 
    75 \paragraph{Internal Fragmentation}
    76 For internal book-keeping, allocators divide raw memory given by OS into chunks, blocks, or lists that can fulfill application's requested size. Allocators use memory given by OS for creating headers, footers etc. to store information about these chunks, blocks, or lists. This increases usage of memory in-addition to the memory requested by application as the allocators need to store their book-keeping information. This extra usage of memory for allocator's own book-keeping is called Internal Fragmentation. Although it cases memory overhead but this overhead is necassary for an allocator's proper funtioning.
    77 
    78 *** FIX ME: Insert a figure of internal fragmentation with explanation
    79 
    80 \paragraph{External Fragmentation}
    81 External fragmentation is the free bits of memory between or around chunks of memory that are currently in-use of the application. Segmentation in memory due to application's usage pattern causes external fragmentation. The memory which is part of external fragmentation is completely free as it is neither used by allocator's internal book-keeping nor by the application. Ideally, an allocator should return a segment of memory back to the OS as soon as application frees it. But, this is not always the case. Allocators get memory from OS in one of the two ways.
    82 
    83 \begin{itemize}
    84 \item
    85 MMap: an allocator can ask OS for whole pages in mmap area. Then, the allocator segments the page internally and fulfills application's request.
    86 \item
    87 Heap: an allocator can ask OS for memory in heap area using system calls such as sbrk. Heap are grows downwards and shrinks upwards.
    88 \begin{itemize}
    89 \item
    90 If an allocator uses mmap area, it can only return extra memory back to OS if the whole page is free i.e. no chunk on the page is in-use of the application. Even if one chunk on the whole page is currently in-use of the application, the allocator has to retain the whole page.
    91 \item
    92 If an allocator uses the heap area, it can only return the continous free memory at the end of the heap area that is currently in allocator's possession as heap area shrinks upwards. If there are free bits of memory in-between chunks of memory that are currently in-use of the application, the allocator can not return these free bits.
    93 
    94 *** FIX ME: Insert a figure of above scenrio with explanation
    95 \item
    96 Even if the entire heap area is free except one small chunk at the end of heap area that is being used by the application, the allocator cannot return the free heap area back to the OS as it is not a continous region at the end of heap area.
    97 
    98 *** FIX ME: Insert a figure of above scenrio with explanation
    99 
    100 \item
    101 Such scenerios cause external fragmentation but it is out of the allocator's control and depend on application's usage pattern.
    102 \end{itemize}
    103 \end{itemize}
    104 
    105 \subsubsection{Internal Memory Management}
    106 Allocators such as je-malloc (FIX ME: insert reference) pro-actively get some memory from the OS and divide it into chunks of certain sizes that can be used in-future to fulfill application's request. This causes memory overhead as these chunks are made before application's request. There is also the possibility that an application may not even request memory of these sizes during their whole life-time.
    107 
    108 *** FIX ME: Insert a figure of above scenrio with explanation
    109 
    110 Allocators such as rp-malloc (FIX ME: insert reference) maintain lists or blocks of sized memory segments that is freed by the application for future use. These lists are maintained without any guarantee that application will even request these sizes again.
    111 
    112 Such tactics are usually used to gain speed as allocator will not have to get raw memory from OS and manage it at the time of application's request but they do cause memory overhead.
    113 
    114 Fragmentation and managed sized chunks of free memory can lead to Heap Blowup as the allocator may not be able to use the fragments or sized free chunks of free memory to fulfill application's requests of other sizes.
    115 
    116 \subsection{Speed}
    117 When it comes to performance evaluation of any piece of software, its runtime is usually the first thing that is evaluated. The same is true for memory allocators but, in case of memory allocators, speed does not only mean the runtime of memory allocator's routines but there are other factors too.
    118 
    119 \subsubsection{Runtime Speed}
    120 Low runtime is the main goal of a memory allocator when it comes it proving its speed. Runtime is the time that it takes for a routine of memory allocator to complete its execution. As mentioned in (FIX ME: refernce to routines' list), there four basic routines that are used in memory allocation. Ideally, each routine of a memory allocator should be fast. Some memory allocator designs use pro-active measures (FIX ME: local refernce) to gain speed when allocating some memory to the application. Some memory allocators do memory allocation faster than memory freeing (FIX ME: graph refernce) while others show similar speed whether memory is allocated or freed.
    121 
    122 \subsubsection{Memory Access Speed}
    123 Runtime speed is not the only speed matrix in memory allocators. The memory that a memory allocator has allocated to the application also needs to be accessible as quick as possible. The application should be able to read/write allocated memory quickly. The allocation method of a memory allocator may introduce some delays when it comes to memory access speed, which is specially important in concurrent applications. Ideally, a memory allocator should allocate all memory on a cache-line to only one thread and no cache-line should be shared among multiple threads. If a memory allocator allocates memory to multple threads on a same cache line, then cache may get invalidated more frequesntly when two different threads running on two different processes will try to read/write the same memory region. On the other hand, if one cache-line is used by only one thread then the cache may get invalidated less frequently. This sharing of one cache-line among multiple threads is called false sharing (FIX ME: cite wasik).
    124 
    125 \paragraph{Active False Sharing}
    126 Active false sharing is the sharing of one cache-line among multiple threads that is caused by memory allocator. It happens when two threads request memory from memory allocator and the allocator allocates memory to both of them on the same cache-line. After that, if the threads are running on different processes who have their own caches and both threads start reading/writing the allocated memory simultanously, their caches will start getting invalidated every time the other thread writes something to the memory. This will cause the application to slow down as the process has to load cache much more frequently.
    127 
    128 *** FIX ME: Insert a figure of above scenrio with explanation
    129 
    130 \paragraph{Passive False Sharing}
    131 Passive false sharing is the kind of false sharing which is caused by the application and not the memory allocator. The memory allocator may preservce passive false sharing in future instead of eradicating it. But, passive false sharing is initiated by the application.
    132 
    133 \subparagraph{Program Induced Passive False Sharing}
    134 Program induced false sharing is completely out of memory allocator's control and is purely caused by the application. When a thread in the application creates multiple objects in the dynamic area and allocator allocates memory for these objects on the same cache-line as the objects are created by the same thread. Passive false sharing will occur if this thread passes one of these objects to another thread but it retains the rest of these objects or it passes some/all of the remaining objects to some third thread(s). Now, one cache-line is shared among multiple threads but it is caused by the application and not the allocator. It is out of allocator's control and has the similar performance impact as Active False Sharing (FIX ME: cite local) if these threads, who are sharing the same cache-line, start reading/writing the given objects simultanously.
    135 
    136 *** FIX ME: Insert a figure of above scenrio 1 with explanation
    137 
    138 *** FIX ME: Insert a figure of above scenrio 2 with explanation
    139 
    140 \subparagraph{Program Induced Allocator Preserved Passive False Sharing}
    141 Program induced allocator preserved passive false sharing is another interesting case of passive false sharing. Both the application and the allocator are partially responsible for it. It starts the same as Program Induced False Sharing (FIX ME: cite local). Once, an application thread has created multiple dynamic objects on the same cache-line and ditributed these objects among multiple threads causing sharing of one cache-line among multiple threads (Program Induced Passive False Sharing). This kind of false sharing occurs when one of these threads, which got the object on the shared cache-line, frees the passed object then re-allocates another object but the allocator returns the same object (on the shared cache-line) that this thread just freed. Although, the application caused the false sharing to happen in the frst place however, to prevent furthur false sharing, the allocator should have returned the new object on some other cache-line which is only shared by the allocating thread. When it comes to performnce impact, this passive false sharing will slow down the application just like any other kind of false sharing if the threads sharing the cache-line start reading/writing the objects simultanously.
    142 
    143 
    144 *** FIX ME: Insert a figure of above scenrio with explanation
    145 
    146 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    147 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    148 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Micro Benchmark Suite
    149 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    150 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    151 
    152 \section{Micro Benchmark Suite}
    153 The aim of micro benchmark suite is to create a set of programs that can evaluate a memory allocator based on the performance matrices described in (FIX ME: local cite). These programs can be taken as a standard to benchmark an allocator's basic goals. These programs give details of an allocator's memory overhead and speed under a certain allocation pattern. The speed of the allocator is benchmarked in different ways. Similarly, false sharing happening in an allocator is also measured in multiple ways. These benchmarks evalute the allocator under a certain allocation pattern which is configurable and can be changed using a few knobs to benchmark observe an allocator's performance under a desired allocation pattern.
    154 
    155 Micro Benchmark Suite benchmarks an allocator's performance by allocating dynamic objects and, then, measuring specifc matrices. The benchmark suite evaluates an allocator with a certain allocation pattern. Bnechmarks have different knobs that can be used to change allocation pattern and evaluate an allocator under desired conditions. These can be set by giving commandline arguments to the benchmark on execution.
    156 
    157 Following is the list of avalable knobs.
    158 
    159 *** FIX ME: Add knobs items after finalize
    160 
    161 \subsection{Memory Benchmark}
    162 Memory benchmark measures memory overhead of an allocator. It allocates a number of dynamic objects. Then, by reading /self/proc/maps, gets the total memory that the allocator has reuested from the OS. Finally, it calculates the memory head by taking the difference between the memory the allocator has requested from the OS and the memory that program has allocated.
    163 *** FIX ME: Insert a figure of above benchmark with description
    164 
    165 \paragraph{Relevant Knobs}
    166 *** FIX ME: Insert Relevant Knobs
    167 
    168 \subsection{Speed Benchmark}
    169 Speed benchmark calculates the runtime speed of an allocator's functions (FIX ME: cite allocator routines). It does by measuring the runtime of allocator routines in two different ways.
    170 
    171 \subsubsection{Speed Time}
    172 The time method does a certain amount of work by calling each routine of the allocator (FIX ME: cite allocator routines) a specific time. It calculates the total time it took to perform this workload. Then, it divides the time it took by the workload and calculates the average time taken by the allocator's routine.
    173 *** FIX ME: Insert a figure of above benchmark with description
    174 
    175 \paragraph{Relevant Knobs}
    176 *** FIX ME: Insert Relevant Knobs
    177 
    178 \subsubsection{Speed Workload}
    179 The worload method uses the opposite approach. It calls the allocator's routines for a specific amount of time and measures how much work was done during that time. Then, similar to the time method, it divides the time by the workload done during that time and calculates the average time taken by the allocator's routine.
    180 *** FIX ME: Insert a figure of above benchmark with description
    181 
    182 \paragraph{Relevant Knobs}
    183 *** FIX ME: Insert Relevant Knobs
    184 
    185 \subsection{Cache Scratch}
    186 Cache Scratch benchmark measures program induced allocator preserved passive false sharing (FIX ME CITE) in an allocator. It does so in two ways.
    187 
    188 \subsubsection{Cache Scratch Time}
    189 Cache Scratch Time allocates dynamic objects. Then, it benchmarks program induced allocator preserved passive false sharing (FIX ME CITE) in an allocator by measuring the time it takes to read/write these objects.
    190 *** FIX ME: Insert a figure of above benchmark with description
    191 
    192 \paragraph{Relevant Knobs}
    193 *** FIX ME: Insert Relevant Knobs
    194 
    195 \subsubsection{Cache Scratch Layout}
    196 Cache Scratch Layout also allocates dynamic objects. Then, it benchmarks program induced allocator preserved passive false sharing (FIX ME CITE) by using heap addresses returned by the allocator. It calculates how many objects were allocated to different threads on the same cache line.
    197 *** FIX ME: Insert a figure of above benchmark with description
    198 
    199 \paragraph{Relevant Knobs}
    200 *** FIX ME: Insert Relevant Knobs
    201 
    202 \subsection{Cache Thrash}
    203 Cache Thrash benchmark measures allocator induced passive false sharing (FIX ME CITE) in an allocator. It also does so in two ways.
    204 
    205 \subsubsection{Cache Thrash Time}
    206 Cache Thrash Time allocates dynamic objects. Then, it benchmarks allocator induced false sharing (FIX ME CITE) in an allocator by measuring the time it takes to read/write these objects.
    207 *** FIX ME: Insert a figure of above benchmark with description
    208 
    209 \paragraph{Relevant Knobs}
    210 *** FIX ME: Insert Relevant Knobs
    211 
    212 \subsubsection{Cache Thrash Layout}
    213 Cache Thrash Layout also allocates dynamic objects. Then, it benchmarks allocator induced false sharing (FIX ME CITE) by using heap addresses returned by the allocator. It calculates how many objects were allocated to different threads on the same cache line.
    214 *** FIX ME: Insert a figure of above benchmark with description
    215 
    216 \paragraph{Relevant Knobs}
    217 *** FIX ME: Insert Relevant Knobs
     88\begin{figure}
     89\centering
     90\includegraphics[width=1\textwidth]{figures/bench-speed.eps}
     91\caption{Benchmark Speed}
     92\label{fig:benchSpeedFig}
     93\end{figure}
     94
     95As laid out in figure \ref{fig:benchSpeedFig}, each chain is measured separately. Each routine in the chain is called for N objects and then
     96 those allocated objects are used when call the next routine in the allocation chain. This way we can measure the
     97 complete latency of memory allocator when multiple routines are chained together e.g. malloc-realloc-free-calloc gives
     98 us the whole picture of the major allocation routines when combined together in a chain.
     99
     100For each chain, time taken is recorded which then can be used to visualize performance of a memory allocator against
     101each chain.
     102
     103Number of worker threads can be adjust using a command-line argument -threadN.
     104
     105\section{Churn Benchmark} Churn benchmark measures the overall runtime speed of an allocator in a multi-threaded
     106 scenerio where each thread extinsevly allocates and frees dynamic memory.
     107
     108\begin{figure}
     109\centering
     110\includegraphics[width=1\textwidth]{figures/bench-churn.eps}
     111\caption{Benchmark Churn}
     112\label{fig:benchChurnFig}
     113\end{figure}
     114
     115Figure \ref{fig:benchChurnFig} illustrates churn benchmark.
     116 This benchmark creates a buffer with M spots and starts K threads. Each thread randomly picks a
     117 spot out of M spots, it frees the object currently at that spot and allocates a new object for that spot. Each thread
     118 repeats this cycle for N times. Main threads measures the total time taken for the whole benchmark and that time is
     119 used to evaluate memory allocator's performance.
     120
     121Only malloc and free are used to allocate and free an object to eliminate any extra cost such as memcpy in realloc etc.
     122Malloc/free allows us to measure latency of memory allocation only without paying any extra cost. Churn simulates a
     123memory intensive program that can be tuned to create different scenerios.
     124
     125Following commandline arguments can be used to tune the benchmark.\\
     126-threadN :  sets number of threads, K\\
     127-cSpots  :  sets number of spots for churn, M\\
     128-objN    :  sets number of objects per thread, N\\
     129-maxS    :  sets max object size\\
     130-minS    :  sets min object size\\
     131-stepS   :  sets object size increment\\
     132-distroS :  sets object size distribution
     133
     134\section{Cache Thrash}\label{sec:benchThrashSec} Cache Thrash benchmark measures allocator induced active false sharing
     135 in an allocator as illustrated in figure \ref{f:AllocatorInducedActiveFalseSharing}.
     136 If memory allocator allocates memory for multiple threads on
     137 same cache line, this can slow down the program performance. If both threads, who share one cache line, frequently
     138 read/write to their object on the cache line concurrently then this will cause cache miss everytime a thread accesse
     139 the object as the other thread might have written something at their memory location on the same cache line.
     140
     141\begin{figure}
     142\centering
     143\includegraphics[width=1\textwidth]{figures/bench-cache-thrash.eps}
     144\caption{Benchmark Allocator Induced Active False Sharing}
     145\label{fig:benchThrashFig}
     146\end{figure}
     147
     148Cache thrash tries to create a scenerio that should lead to allocator induced false sharing if the underlying memory
     149allocator is allocating dynamic memory to multiple threads on same cache lines. Ideally, a memory allocator should
     150distance dynamic memory region of one thread from other threads'. Having multiple threads allocating small objects
     151simultanously should cause the memory allocator to allocate objects for multiple objects on the same cache line if its
     152not distancing the memory among different threads.
     153
     154Figure \ref{fig:benchThrashFig} lays out flow of the cache thrash benchmark.
     155 It creates K worker threads. Each worker thread allocates an object and intensively read/write
     156 it for M times to invalidate cache lines frequently to slow down other threads who might be sharing this cache line
     157 with it. Each thread repeats this for N times. Main thread measures the total time taken to for all worker threads to
     158 complete. Worker threads sharing cahche lines with each other will take longer.
     159
     160Different cache access scenerios can be created using the following commandline arguments.\\
     161-threadN :  sets number of threads, K\\
     162-cacheIt :  iterations for cache benchmark, N\\
     163-cacheRep:  repetations for cache benchmark, M\\
     164-cacheObj:  object size for cache benchmark
     165
     166\section{Cache Scratch} Cache Scratch benchmark measures allocator induced passive false sharing in an allocator. An
     167 allocator can unintentionally induce false sharing depending upon its management of the freed objects as described in
     168 figure \ref{f:AllocatorInducedPassiveFalseSharing}. If a thread A allocates multiple objects together then they will be
     169  possibly allocated on the same cache line by the memory allocator. If the thread now passes this object to another
     170  thread B then the two of them will sharing the same cache line but this scenerio is not induced by the allocator.
     171  Instead, the program induced this situation. Now it might be possible that if thread B frees this object and then
     172  allocate an object of the same size then the allocator may return the same object which is on a cache line shared
     173  with thread A. Now this false sharing is being caused by the memory allocator although it was started by the
     174  program.
     175
     176\begin{figure}
     177\centering
     178\includegraphics[width=1\textwidth]{figures/bench-cache-scratch.eps}
     179\caption{Benchmark Program Induced Passive False Sharing}
     180\label{fig:benchScratchFig}
     181\end{figure}
     182
     183Cache scratch main thread induces false sharing and creates a scenerio that should make memory allocator preserve the
     184 program-induced false sharing if it does not retur a freed object to its owner thread and, instead, re-uses it
     185 instantly. An alloator using object ownership, as described in section \ref{s:Ownership}, would be less susceptible to allocator induced passive
     186 false sharing. If the object is returned to the thread who owns it or originally allocated it then the thread B will
     187 get a new object that will be less likely to be on the same cache line as thread A.
     188
     189As in figure \ref{fig:benchScratchFig}, cache Scratch allocates K dynamic objects together, one for each of the K worker threads,
     190 possibly causing memory allocator to allocate these objects on the same cache-line. Then it create K worker threads and passes
     191 an object from the K allocated objects to each of the K threads. Each worker thread frees the object passed by the main thread.
     192 Then, it allocates an object and reads/writes it repetitively for M times causing frequent cache invalidations. Each worker
     193 repeats this for N times.
     194
     195Each thread allocating an object after freeing the original object passed by the main thread should cause the memory
     196allocator to return the same object that was initially allocated by the main thread if the allocator did not return the
     197intial object bakc to its owner (main thread). Then, intensive read/write on the shared cache line by multiple threads
     198should slow down worker threads due to to high cache invalidations and misses. Main thread measures the total time
     199taken for all the workers to complete.
     200
     201Similar to bechmark cache thrash in section \ref{sec:benchThrashSec}, different cache access scenerios can be created using the following commandline arguments.\\
     202-threadN :  sets number of threads, K\\
     203-cacheIt :  iterations for cache benchmark, N\\
     204-cacheRep:  repetations for cache benchmark, M\\
     205-cacheObj:  object size for cache benchmark

  • doc/theses/mubeen_zulfiqar_MMath/performance.tex

    rbb7c77d r2e9b59b  
    11\chapter{Performance}
    22\label{c:Performance}
    3 
    4 \noindent
    5 ====================
    6 
    7 Writing Points:
    8 \begin{itemize}
    9 \item
    10 Machine Specification
    11 \item
    12 Allocators and their details
    13 \item
    14 Benchmarks and their details
    15 \item
    16 Results
    17 \end{itemize}
    18 
    19 \noindent
    20 ====================
    213
    224\section{Machine Specification}
    … …  
    257\begin{itemize}
    268\item
    27 AMD EPYC 7662, 64-core socket $\times$ 2, 2.0 GHz
     9{\bf Nasus} AMD EPYC 7662, 64-core socket $\times$ 2, 2.0 GHz, GCC version 9.3.0
    2810\item
    29 Huawei ARM TaiShan 2280 V2 Kunpeng 920, 24-core socket $\times$ 4, 2.6 GHz
    30 \item
    31 Intel Xeon Gold 5220R, 48-core socket $\times$ 2, 2.20GHz
     11{\bf Algol} Huawei ARM TaiShan 2280 V2 Kunpeng 920, 24-core socket $\times$ 4, 2.6 GHz, GCC version 9.4.0
    3212\end{itemize}
    3313
    3414
    35 \section{Existing Memory Allocators}
     15\section{Existing Memory Allocators}\label{sec:curAllocatorSec}
    3616With dynamic allocation being an important feature of C, there are many stand-alone memory allocators that have been designed for different purposes. For this thesis, we chose 7 of the most popular and widely used memory allocators.
    3717
    38 \paragraph{dlmalloc}
    39 dlmalloc (FIX ME: cite allocator) is a thread-safe allocator that is single threaded and single heap. dlmalloc maintains free-lists of different sizes to store freed dynamic memory. (FIX ME: cite wasik)
    40 
    41 \paragraph{hoard}
     18\subsection{dlmalloc}
     19dlmalloc (FIX ME: cite allocator with download link) is a thread-safe allocator that is single threaded and single heap. dlmalloc maintains free-lists of different sizes to store freed dynamic memory. (FIX ME: cite wasik)
     20\\
     21\\
     22{\bf Version:} 2.8.6\\
     23{\bf Configuration:} Compiled with pre-processor USE\_LOCKS.\\
     24{\bf Compilation command:}\\
     25cc -g3 -O3 -Wall -Wextra -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-free -fPIC -shared -DUSE\_LOCKS -o libdlmalloc.so malloc-2.8.6.c
     26
     27\subsection{hoard}
    4228Hoard (FIX ME: cite allocator) is a thread-safe allocator that is multi-threaded and using a heap layer framework. It has per-thread heaps that have thread-local free-lists, and a global shared heap. (FIX ME: cite wasik)
    43 
    44 \paragraph{jemalloc}
     29\\
     30\\
     31{\bf Version:} 3.13\\
     32{\bf Configuration:} Compiled with hoard's default configurations and Makefile.\\
     33{\bf Compilation command:}\\
     34make all
     35
     36\subsection{jemalloc}
    4537jemalloc (FIX ME: cite allocator) is a thread-safe allocator that uses multiple arenas. Each thread is assigned an arena. Each arena has chunks that contain contagious memory regions of same size. An arena has multiple chunks that contain regions of multiple sizes.
    46 
    47 \paragraph{ptmalloc}
    48 ptmalloc (FIX ME: cite allocator) is a modification of dlmalloc. It is a thread-safe multi-threaded memory allocator that uses multiple heaps. ptmalloc heap has similar design to dlmalloc's heap.
    49 
    50 \paragraph{rpmalloc}
     38\\
     39\\
     40{\bf Version:} 5.2.1\\
     41{\bf Configuration:} Compiled with jemalloc's default configurations and Makefile.\\
     42{\bf Compilation command:}\\
     43./autogen.sh\\
     44./configure\\
     45make\\
     46make install
     47
     48\subsection{pt3malloc}
     49pt3malloc (FIX ME: cite allocator) is a modification of dlmalloc. It is a thread-safe multi-threaded memory allocator that uses multiple heaps. pt3malloc heap has similar design to dlmalloc's heap.
     50\\
     51\\
     52{\bf Version:} 1.8\\
     53{\bf Configuration:} Compiled with pt3malloc's Makefile using option "linux-shared".\\
     54{\bf Compilation command:}\\
     55make linux-shared
     56
     57\subsection{rpmalloc}
    5158rpmalloc (FIX ME: cite allocator) is a thread-safe allocator that is multi-threaded and uses per-thread heap. Each heap has multiple size-classes and each size-class contains memory regions of the relevant size.
    52 
    53 \paragraph{tbb malloc}
     59\\
     60\\
     61{\bf Version:} 1.4.1\\
     62{\bf Configuration:} Compiled with rpmalloc's default configurations and ninja build system.\\
     63{\bf Compilation command:}\\
     64python3 configure.py\\
     65ninja
     66
     67\subsection{tbb malloc}
    5468tbb malloc (FIX ME: cite allocator) is a thread-safe allocator that is multi-threaded and uses private heap for each thread. Each private-heap has multiple bins of different sizes. Each bin contains free regions of the same size.
    55 
    56 \paragraph{tc malloc}
    57 tcmalloc (FIX ME: cite allocator) is a thread-safe allocator. It uses per-thread cache to store free objects that prevents contention on shared resources in multi-threaded application. A central free-list is used to refill per-thread cache when it gets empty.
    58 
    59 
    60 \section{Memory Allocators}
    61 For these experiments, we used 7 memory allocators excluding our standalone memory allocator uHeapLmmm.
    62 
    63 \begin{tabularx}{0.8\textwidth} {
    64         | >{\raggedright\arraybackslash}X
    65         | >{\centering\arraybackslash}X
    66         | >{\raggedleft\arraybackslash}X |
    67 }
    68 \hline
    69 Memory Allocator & Version     & Configurations \\
    70 \hline
    71 dl               &             &  \\
    72 \hline
    73 hoard            &             &  \\
    74 \hline
    75 je               &             &  \\
    76 \hline
    77 pt3              &             &  \\
    78 \hline
    79 rp               &             &  \\
    80 \hline
    81 tbb              &             &  \\
    82 \hline
    83 tc               &             &  \\
    84 \end{tabularx}
    85 
    86 %(FIX ME: complete table)
     69\\
     70\\
     71{\bf Version:} intel tbb 2020 update 2, tbb\_interface\_version == 11102\\
     72{\bf Configuration:} Compiled with tbbmalloc's default configurations and Makefile.\\
     73{\bf Compilation command:}\\
     74make
    8775
    8876\section{Experiment Environment}
    89 We conducted these experiments ... (FIX ME: what machine and which specifications to add).
    90 
    91 We used our micro becnhmark suite (FIX ME: cite mbench) to evaluate other memory allocators (FIX ME: cite above memory allocators) and our uHeapLmmm.
     77We used our micro becnhmark suite (FIX ME: cite mbench) to evaluate these memory allocators \ref{sec:curAllocatorSec} and our own memory allocator uHeap \ref{sec:allocatorSec}.
    9278
    9379\section{Results}
     80FIX ME: add experiment, knobs, graphs, description+analysis
     81
     82%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     83%% CHURN
     84%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     85
     86\subsection{Churn Benchmark}
     87
     88Churn benchmark tested memory allocators for speed under intensive dynamic memory usage.
     89
     90This experiment was run with following configurations:
     91
     92-maxS            : 500
     93
     94-minS            : 50
     95
     96-stepS           : 50
     97
     98-distroS         : fisher
     99
     100-objN            : 100000
     101
     102-cSpots          : 16
     103
     104-threadN         : \{ 1, 2, 4, 8, 16 \} *
     105
     106* Each allocator was tested for its performance across different number of threads. Experiment was repeated for each allocator for 1, 2, 4, 8, and 16 threads by setting the configuration -threadN.
     107
     108Results are shown in figure \ref{fig:churn} for both algol and nasus.
     109X-axis shows number of threads. Each allocator's performance for each thread is shown in different colors.
     110Y-axis shows the total time experiment took to finish.
     111
     112\begin{figure}
     113\centering
     114    \subfigure[Algol]{ \includegraphics[width=0.9\textwidth]{evaluations/algol-perf-eps/churn} }
     115    \subfigure[Nasus]{ \includegraphics[width=0.9\textwidth]{evaluations/nasus-perf-eps/churn} }
     116\caption{Churn}
     117\label{fig:churn}
     118\end{figure}
     119
     120%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     121%% THRASH
     122%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     123
     124\subsection{Cache Thrash}
     125
     126Thrash benchmark tested memory allocators for active false sharing.
     127
     128This experiment was run with following configurations:
     129
     130-cacheIt        : 1000
     131
     132-cacheRep       : 1000000
     133
     134-cacheObj       : 1
     135
     136-threadN        : \{ 1, 2, 4, 8, 16 \} *
     137
     138* Each allocator was tested for its performance across different number of threads. Experiment was repeated for each allocator for 1, 2, 4, 8, and 16 threads by setting the configuration -threadN.
     139
     140Results are shown in figure \ref{fig:cacheThrash} for both algol and nasus.
     141X-axis shows number of threads. Each allocator's performance for each thread is shown in different colors.
     142Y-axis shows the total time experiment took to finish.
     143
     144\begin{figure}
     145\centering
     146    \subfigure[Algol]{ \includegraphics[width=0.9\textwidth]{evaluations/algol-perf-eps/cache-time-0-thrash} }
     147    \subfigure[Nasus]{ \includegraphics[width=0.9\textwidth]{evaluations/nasus-perf-eps/cache-time-0-thrash} }
     148\caption{Cache Thrash}
     149\label{fig:cacheThrash}
     150\end{figure}
     151
     152%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     153%% SCRATCH
     154%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     155
     156\subsection{Cache Scratch}
     157
     158Scratch benchmark tested memory allocators for program induced allocator preserved passive false sharing.
     159
     160This experiment was run with following configurations:
     161
     162-cacheIt        : 1000
     163
     164-cacheRep       : 1000000
     165
     166-cacheObj       : 1
     167
     168-threadN        : \{ 1, 2, 4, 8, 16 \} *
     169
     170* Each allocator was tested for its performance across different number of threads. Experiment was repeated for each allocator for 1, 2, 4, 8, and 16 threads by setting the configuration -threadN.
     171
     172Results are shown in figure \ref{fig:cacheScratch} for both algol and nasus.
     173X-axis shows number of threads. Each allocator's performance for each thread is shown in different colors.
     174Y-axis shows the total time experiment took to finish.
     175
     176\begin{figure}
     177\centering
     178    \subfigure[Algol]{ \includegraphics[width=0.9\textwidth]{evaluations/algol-perf-eps/cache-time-0-scratch} }
     179    \subfigure[Nasus]{ \includegraphics[width=0.9\textwidth]{evaluations/nasus-perf-eps/cache-time-0-scratch} }
     180\caption{Cache Scratch}
     181\label{fig:cacheScratch}
     182\end{figure}
     183
     184%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     185%% SPEED
     186%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     187
     188\subsection{Speed Benchmark}
     189
     190Speed benchmark tested memory allocators for program induced allocator preserved passive false sharing.
     191
     192This experiment was run with following configurations:
     193
     194-threadN :  sets number of threads, K\\
     195-cSpots  :  sets number of spots for churn, M\\
     196-objN    :  sets number of objects per thread, N\\
     197-maxS    :  sets max object size\\
     198-minS    :  sets min object size\\
     199-stepS   :  sets object size increment\\
     200-distroS :  sets object size distribution
     201
     202%speed-1-malloc-null.eps
     203\begin{figure}
     204\centering
     205\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-1-malloc-null}
     206\caption{speed-1-malloc-null}
     207\label{fig:speed-1-malloc-null}
     208\end{figure}
     209
     210%speed-2-free-null.eps
     211\begin{figure}
     212\centering
     213\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-2-free-null}
     214\caption{speed-2-free-null}
     215\label{fig:speed-2-free-null}
     216\end{figure}
     217
     218%speed-3-malloc.eps
     219\begin{figure}
     220\centering
     221\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-3-malloc}
     222\caption{speed-3-malloc}
     223\label{fig:speed-3-malloc}
     224\end{figure}
     225
     226%speed-4-realloc.eps
     227\begin{figure}
     228\centering
     229\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-4-realloc}
     230\caption{speed-4-realloc}
     231\label{fig:speed-4-realloc}
     232\end{figure}
     233
     234%speed-5-free.eps
     235\begin{figure}
     236\centering
     237\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-5-free}
     238\caption{speed-5-free}
     239\label{fig:speed-5-free}
     240\end{figure}
     241
     242%speed-6-calloc.eps
     243\begin{figure}
     244\centering
     245\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-6-calloc}
     246\caption{speed-6-calloc}
     247\label{fig:speed-6-calloc}
     248\end{figure}
     249
     250%speed-7-malloc-free.eps
     251\begin{figure}
     252\centering
     253\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-7-malloc-free}
     254\caption{speed-7-malloc-free}
     255\label{fig:speed-7-malloc-free}
     256\end{figure}
     257
     258%speed-8-realloc-free.eps
     259\begin{figure}
     260\centering
     261\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-8-realloc-free}
     262\caption{speed-8-realloc-free}
     263\label{fig:speed-8-realloc-free}
     264\end{figure}
     265
     266%speed-9-calloc-free.eps
     267\begin{figure}
     268\centering
     269\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-9-calloc-free}
     270\caption{speed-9-calloc-free}
     271\label{fig:speed-9-calloc-free}
     272\end{figure}
     273
     274%speed-10-malloc-realloc.eps
     275\begin{figure}
     276\centering
     277\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-10-malloc-realloc}
     278\caption{speed-10-malloc-realloc}
     279\label{fig:speed-10-malloc-realloc}
     280\end{figure}
     281
     282%speed-11-calloc-realloc.eps
     283\begin{figure}
     284\centering
     285\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-11-calloc-realloc}
     286\caption{speed-11-calloc-realloc}
     287\label{fig:speed-11-calloc-realloc}
     288\end{figure}
     289
     290%speed-12-malloc-realloc-free.eps
     291\begin{figure}
     292\centering
     293\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-12-malloc-realloc-free}
     294\caption{speed-12-malloc-realloc-free}
     295\label{fig:speed-12-malloc-realloc-free}
     296\end{figure}
     297
     298%speed-13-calloc-realloc-free.eps
     299\begin{figure}
     300\centering
     301\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-13-calloc-realloc-free}
     302\caption{speed-13-calloc-realloc-free}
     303\label{fig:speed-13-calloc-realloc-free}
     304\end{figure}
     305
     306%speed-14-{m,c,re}alloc-free.eps
     307\begin{figure}
     308\centering
     309\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/speed-14-{m,c,re}alloc-free}
     310\caption{speed-14-{m,c,re}alloc-free}
     311\label{fig:speed-14-{m,c,re}alloc-free}
     312\end{figure}
     313
     314%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     315%% MEMORY
     316%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    94317
    95318\subsection{Memory Benchmark}
    96 FIX ME: add experiment, knobs, graphs, and description
    97 
    98 \subsection{Speed Benchmark}
    99 FIX ME: add experiment, knobs, graphs, and description
    100 
    101 \subsubsection{Speed Time}
    102 FIX ME: add experiment, knobs, graphs, and description
    103 
    104 \subsubsection{Speed Workload}
    105 FIX ME: add experiment, knobs, graphs, and description
    106 
    107 \subsection{Cache Scratch}
    108 FIX ME: add experiment, knobs, graphs, and description
    109 
    110 \subsubsection{Cache Scratch Time}
    111 FIX ME: add experiment, knobs, graphs, and description
    112 
    113 \subsubsection{Cache Scratch Layout}
    114 FIX ME: add experiment, knobs, graphs, and description
    115 
    116 \subsection{Cache Thrash}
    117 FIX ME: add experiment, knobs, graphs, and description
    118 
    119 \subsubsection{Cache Thrash Time}
    120 FIX ME: add experiment, knobs, graphs, and description
    121 
    122 \subsubsection{Cache Thrash Layout}
    123 FIX ME: add experiment, knobs, graphs, and description
     319
     320%mem-1-prod-1-cons-100-cfa.eps
     321\begin{figure}
     322\centering
     323\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-1-prod-1-cons-100-cfa}
     324\caption{mem-1-prod-1-cons-100-cfa}
     325\label{fig:mem-1-prod-1-cons-100-cfa}
     326\end{figure}
     327
     328%mem-1-prod-1-cons-100-dl.eps
     329\begin{figure}
     330\centering
     331\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-1-prod-1-cons-100-dl}
     332\caption{mem-1-prod-1-cons-100-dl}
     333\label{fig:mem-1-prod-1-cons-100-dl}
     334\end{figure}
     335
     336%mem-1-prod-1-cons-100-glc.eps
     337\begin{figure}
     338\centering
     339\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-1-prod-1-cons-100-glc}
     340\caption{mem-1-prod-1-cons-100-glc}
     341\label{fig:mem-1-prod-1-cons-100-glc}
     342\end{figure}
     343
     344%mem-1-prod-1-cons-100-hrd.eps
     345\begin{figure}
     346\centering
     347\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-1-prod-1-cons-100-hrd}
     348\caption{mem-1-prod-1-cons-100-hrd}
     349\label{fig:mem-1-prod-1-cons-100-hrd}
     350\end{figure}
     351
     352%mem-1-prod-1-cons-100-je.eps
     353\begin{figure}
     354\centering
     355\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-1-prod-1-cons-100-je}
     356\caption{mem-1-prod-1-cons-100-je}
     357\label{fig:mem-1-prod-1-cons-100-je}
     358\end{figure}
     359
     360%mem-1-prod-1-cons-100-pt3.eps
     361\begin{figure}
     362\centering
     363\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-1-prod-1-cons-100-pt3}
     364\caption{mem-1-prod-1-cons-100-pt3}
     365\label{fig:mem-1-prod-1-cons-100-pt3}
     366\end{figure}
     367
     368%mem-1-prod-1-cons-100-rp.eps
     369\begin{figure}
     370\centering
     371\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-1-prod-1-cons-100-rp}
     372\caption{mem-1-prod-1-cons-100-rp}
     373\label{fig:mem-1-prod-1-cons-100-rp}
     374\end{figure}
     375
     376%mem-1-prod-1-cons-100-tbb.eps
     377\begin{figure}
     378\centering
     379\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-1-prod-1-cons-100-tbb}
     380\caption{mem-1-prod-1-cons-100-tbb}
     381\label{fig:mem-1-prod-1-cons-100-tbb}
     382\end{figure}
     383
     384%mem-4-prod-4-cons-100-cfa.eps
     385\begin{figure}
     386\centering
     387\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-4-prod-4-cons-100-cfa}
     388\caption{mem-4-prod-4-cons-100-cfa}
     389\label{fig:mem-4-prod-4-cons-100-cfa}
     390\end{figure}
     391
     392%mem-4-prod-4-cons-100-dl.eps
     393\begin{figure}
     394\centering
     395\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-4-prod-4-cons-100-dl}
     396\caption{mem-4-prod-4-cons-100-dl}
     397\label{fig:mem-4-prod-4-cons-100-dl}
     398\end{figure}
     399
     400%mem-4-prod-4-cons-100-glc.eps
     401\begin{figure}
     402\centering
     403\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-4-prod-4-cons-100-glc}
     404\caption{mem-4-prod-4-cons-100-glc}
     405\label{fig:mem-4-prod-4-cons-100-glc}
     406\end{figure}
     407
     408%mem-4-prod-4-cons-100-hrd.eps
     409\begin{figure}
     410\centering
     411\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-4-prod-4-cons-100-hrd}
     412\caption{mem-4-prod-4-cons-100-hrd}
     413\label{fig:mem-4-prod-4-cons-100-hrd}
     414\end{figure}
     415
     416%mem-4-prod-4-cons-100-je.eps
     417\begin{figure}
     418\centering
     419\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-4-prod-4-cons-100-je}
     420\caption{mem-4-prod-4-cons-100-je}
     421\label{fig:mem-4-prod-4-cons-100-je}
     422\end{figure}
     423
     424%mem-4-prod-4-cons-100-pt3.eps
     425\begin{figure}
     426\centering
     427\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-4-prod-4-cons-100-pt3}
     428\caption{mem-4-prod-4-cons-100-pt3}
     429\label{fig:mem-4-prod-4-cons-100-pt3}
     430\end{figure}
     431
     432%mem-4-prod-4-cons-100-rp.eps
     433\begin{figure}
     434\centering
     435\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-4-prod-4-cons-100-rp}
     436\caption{mem-4-prod-4-cons-100-rp}
     437\label{fig:mem-4-prod-4-cons-100-rp}
     438\end{figure}
     439
     440%mem-4-prod-4-cons-100-tbb.eps
     441\begin{figure}
     442\centering
     443\includegraphics[width=1\textwidth]{evaluations/nasus-perf-eps/mem-4-prod-4-cons-100-tbb}
     444\caption{mem-4-prod-4-cons-100-tbb}
     445\label{fig:mem-4-prod-4-cons-100-tbb}
     446\end{figure}
Note: See TracChangeset for help on using the changeset viewer.