io_comparison on remote machine

Author: davide-marchi

Makefile

@@ -29,6 +29,8 @@ SRCS := sequential.cpp \
         openmp.cpp     \
         fastflow.cpp   \
         io_comparison.cpp \
+        sequential_onlymmap.cpp \
+        sequential_nommap.cpp \
         # mpi_omp.cpp    \
         # mpi_ff.cpp
 

io_comparison.cpp

@@ -37,9 +37,9 @@ static std::string generate_unsorted_file_streaming(std::size_t    total_n,
         std::exit(1);
     }
 
-    // I/O buffer (1 MiB), flush when >= 512 KiB
-    constexpr std::size_t IO_BUF_SZ    = 1 << 20;
-    constexpr std::size_t FLUSH_THRESH = 512 << 10; // 512 KiB
+    // I/O buffer (1 GiB), flush when >= 512 MiB
+    constexpr std::size_t IO_BUF_SZ    = 1 << 30;
+    constexpr std::size_t FLUSH_THRESH = 512 << 20; // 512 MiB
     std::vector<char> io_buf;
     io_buf.reserve(IO_BUF_SZ);
 
@@ -201,7 +201,7 @@ static std::string generate_unsorted_file_setvbuf(std::size_t total_n,
 
     // 5) tune the stdio buffer to 512 KiB
     std::vector<char> io_buf(512 << 10);
-    if (std::setvbuf(f, nullptr, _IOFBF, 512 << 10) != 0) {
+    if (std::setvbuf(f, nullptr, _IOFBF, stats.st_blksize) != 0) {
         std::perror("setvbuf");
         std::fclose(f);
         std::exit(1);

logs/output_comparison.txt

@@ -0,0 +1,59 @@
+Generated “files/a_unsorted_10000000_256.bin”.
+[stream_write        ]  33634.798 ms
+Streaming-generating 10000000 records into “files/b_unsorted_10000000_256.bin”…
+Unsorted file ready: “files/b_unsorted_10000000_256.bin”.
+[stdio_fwrite        ]  27382.517 ms
+BUFSIZ is 8192, but optimal block size is 1048576
+Streaming-generating 10000000 records into "files/c_unsorted_10000000_256.bin"…
+Unsorted file ready: "files/c_unsorted_10000000_256.bin".
+[setvbuf_fwrite      ]  21690.073 ms
+[generate_arrays     ]    405.065 ms
+[open_fallocate      ]     25.671 ms
+[mmap                ]      0.026 ms
+[generate_records    ]  14824.201 ms
+[teardown            ]  27418.757 ms
+Generated “files/d_unsorted_10000000_256.bin” (1440222028 bytes).
+[mmap_fallocate      ]  42674.708 ms
+Streaming-generating 10000000 records into "files/e_unsorted_10000000_256.bin" with O_DIRECT…
+Unsorted file ready: "files/e_unsorted_10000000_256.bin".
+[odirect_write       ]  89396.495 ms
+
+Generated “files/a_unsorted_10000000_256.bin”.
+[stream_write        ] 108758.485 ms
+Streaming-generating 10000000 records into “files/b_unsorted_10000000_256.bin”…
+Unsorted file ready: “files/b_unsorted_10000000_256.bin”.
+[stdio_fwrite        ]  20413.666 ms
+BUFSIZ is 8192, but optimal block size is 1048576
+Streaming-generating 10000000 records into "files/c_unsorted_10000000_256.bin"…
+Unsorted file ready: "files/c_unsorted_10000000_256.bin".
+[setvbuf_fwrite      ]  47973.324 ms
+[generate_arrays     ]    407.878 ms
+[open_fallocate      ]    148.006 ms
+[mmap                ]      0.040 ms
+[generate_records    ]  14975.573 ms
+[teardown            ]  15495.449 ms
+Generated “files/d_unsorted_10000000_256.bin” (1440222028 bytes).
+[mmap_fallocate      ]  31027.866 ms
+Streaming-generating 10000000 records into "files/e_unsorted_10000000_256.bin" with O_DIRECT…
+Unsorted file ready: "files/e_unsorted_10000000_256.bin".
+[odirect_write       ]  39236.698 ms
+
+Generated “files/a_unsorted_10000000_256.bin”.
+[stream_write        ]  48583.323 ms
+Streaming-generating 10000000 records into “files/b_unsorted_10000000_256.bin”…
+Unsorted file ready: “files/b_unsorted_10000000_256.bin”.
+[stdio_fwrite        ] 105960.016 ms
+BUFSIZ is 8192, but optimal block size is 1048576
+Streaming-generating 10000000 records into "files/c_unsorted_10000000_256.bin"…
+Unsorted file ready: "files/c_unsorted_10000000_256.bin".
+[setvbuf_fwrite      ]  39650.644 ms
+[generate_arrays     ]    405.603 ms
+[open_fallocate      ]      2.253 ms
+[mmap                ]      0.025 ms
+[generate_records    ]  14830.265 ms
+[teardown            ]   6506.072 ms
+Generated “files/d_unsorted_10000000_256.bin” (1440222028 bytes).
+[mmap_fallocate      ]  21744.966 ms
+Streaming-generating 10000000 records into "files/e_unsorted_10000000_256.bin" with O_DIRECT…
+Unsorted file ready: "files/e_unsorted_10000000_256.bin".
+[odirect_write       ]  42179.944 ms

logs/output_onlymmap.txt

@@ -0,0 +1,16 @@
+srun bin/sequential_onlymmap -n 100000000 -p 8
+[generate_arrays     ]   4018.134 ms
+[open_truncate       ]     81.273 ms
+[mmap                ]      0.026 ms
+[generate_records    ]  11999.642 ms
+[teardown            ] 185260.842 ms
+Generated “files/unsorted_100000000_8.bin” (2000000000 bytes).
+[generate_unsorted_file] 201364.572 ms
+[build_index         ]   1068.235 ms
+[sort_records        ]  12470.667 ms
+[open_and_mmap_output]     25.092 ms
+[rewrite_sorted      ]  30734.122 ms
+Verifying output…
+File is sorted.
+[check_if_sorted_mmap]    476.212 ms
+

sequential.cpp

@@ -25,7 +25,7 @@ static std::string generate_unsorted_file_streaming(std::size_t    total_n,
     namespace fs = std::filesystem;
     fs::create_directories("files");
 
-    std::string path = "files/a_unsorted_"
+    std::string path = "files/unsorted_"
                      + std::to_string(total_n) + "_"
                      + std::to_string(payload_max) + ".bin";
 
@@ -41,9 +41,9 @@ static std::string generate_unsorted_file_streaming(std::size_t    total_n,
         std::exit(1);
     }
 
-    // I/O buffer (1 MiB), flush when >= 512 KiB
-    constexpr std::size_t IO_BUF_SZ    = 1 << 20;
-    constexpr std::size_t FLUSH_THRESH = 512 << 10; // 512 KiB
+    // I/O buffer (1 GiB), flush when >= 512 MiB
+    constexpr std::size_t IO_BUF_SZ    = 1 << 30;
+    constexpr std::size_t FLUSH_THRESH = 512 << 20; // 512 MiB
     std::vector<char> io_buf;
     io_buf.reserve(IO_BUF_SZ);
 

todo.md

@@ -17,4 +17,27 @@ Fastflow
         - would remove global variable but would reuire to change the merge_record()
 
 
-Devo cancellare i file ordinati dopo il controllo?
+Devo cancellare i file ordinati dopo il controllo?
+
+
+----
+
+Option bottom-upp (per svrapporre generazione index e ordinamento):
+
+base-case = 10'000
+
+1. leggo il file
+
+2. Ogni struct letta controllo:
+
+    (Sono degli if all'interno di un for, NON uno switch!)
+
+    - Se la struct è la numero 2^0 * base case: Ordino gli ultimi base-case con sort()
+
+    - Se la struct è la numero 2^1 * base case: Mergio gli ultimi 2 array di dim. 2^0 * base-case
+
+    - Se la struct è la numero 2^2 * base case: Mergio glu ultimi 2 array di dim. 2^1 * base case
+
+    - ...
+
+    Caso in cui la struct letta è l'ultima?