The point of this repo is to get an intuition about different IO models. Comparing across languages isn't for benchmark wars it's more a check for correctness. The relative difference between IO models should be similar across language.
I am running these tests on a dedicated bare metal instance, OVH Rise-1.
- RAM: 64 GB DDR4 ECC 2,133 MHz
- Disk: 2x450 GB SSD NVMe in Soft RAID
- Processor: Intel Xeon E3-1230v6 - 4c/8t - 3.5 GHz/3.9 GHz
uname --kernel-release: 6.3.8-100.fc37.x86_64
Each implementation (other than dd) produces a CSV of results. Use
the following DuckDB command to analyze it.
$ duckdb -c "
SELECT
column0 AS method,
AVG(column1::DOUBLE) || 's' avg_time,
FORMAT_BYTES(AVG(column2::DOUBLE)::BIGINT) || '/s' AS avg_throughput
FROM 'out.csv'
GROUP BY column0
ORDER BY AVG(column1::DOUBLE) ASC"
$ dd if=/dev/zero of=test.bin bs=4k count=1M
1048576+0 records in
1048576+0 records out
4294967296 bytes (4.3 GB, 4.0 GiB) copied, 3.09765 s, 1.4 GB/s
This version reads up to N entries but then blocks until all N entries complete. This is not ideal. But I'm not sure Iceber/iouring-go supports anything else.
First:
$ go run main.go | tee out.csv
Then run the DuckDB command from above:
┌────────────────────────────────────────────┬─────────────────────┬────────────────┐
│ method │ avg_time │ avg_throughput │
│ varchar │ varchar │ varchar │
├────────────────────────────────────────────┼─────────────────────┼────────────────┤
│ 1_goroutines_pwrite │ 0.7111268999999999s │ 1.5GB/s │
│ blocking │ 0.7128968s │ 1.5GB/s │
│ 1_goroutines_io_uring_pwrite_128_entries │ 1.0402713s │ 1.0GB/s │
│ 10_goroutines_pwrite │ 1.111215s │ 966.2MB/s │
│ 100_goroutines_io_uring_pwrite_128_entries │ 1.3004915000000001s │ 825.6MB/s │
│ 100_goroutines_io_uring_pwrite_1_entries │ 1.5118257s │ 710.2MB/s │
│ 10_goroutines_io_uring_pwrite_128_entries │ 1.5322980999999998s │ 771.6MB/s │
│ 10_goroutines_io_uring_pwrite_1_entries │ 1.6577722000000001s │ 648.1MB/s │
│ 1_goroutines_io_uring_pwrite_1_entries │ 4.705483s │ 228.2MB/s │
└────────────────────────────────────────────┴─────────────────────┴────────────────┘
Unlike the Go implementation, this version does not batch only/always N entries at a time. It starts out batching N entries at a time but does not block waiting for all N to complete. Instead it just reads what has completed and keeps trying to add more entry batches.
First:
$ zig build-exe main.zig
$ ./main | tee out.csv
Then run the DuckDB command from above:
┌────────────────────────────────────────┬─────────────────────┬────────────────┐
│ method │ avg_time │ avg_throughput │
│ varchar │ varchar │ varchar │
├────────────────────────────────────────┼─────────────────────┼────────────────┤
│ 1_threads_iouring_pwrite_128_entries │ 0.6080365773999998s │ 1.7GB/s │
│ 1_threads_iouring_pwrite_1_entries │ 0.6259650676999999s │ 1.7GB/s │
│ blocking │ 0.6740227804s │ 1.5GB/s │
│ 1_threads_pwrite │ 0.6846085126999999s │ 1.5GB/s │
│ 10_threads_pwrite │ 1.1549885629000003s │ 929.8MB/s │
│ 10_threads_iouring_pwrite_1_entries │ 2.4174379148s │ 445.7MB/s │
│ 10_threads_iouring_pwrite_128_entries │ 2.4178504731s │ 445.8MB/s │
│ 100_threads_iouring_pwrite_128_entries │ 3.6317807736s │ 296.6MB/s │
│ 100_threads_iouring_pwrite_1_entries │ 3.7681755905000003s │ 287.7MB/s │
└────────────────────────────────────────┴─────────────────────┴────────────────┘
First:
$ python3 main.py | tee out.csv
Then run the DuckDB command from above:
┌──────────┬────────────┬────────────────┐
│ method │ avg_time │ avg_throughput │
│ varchar │ varchar │ varchar │
├──────────┼────────────┼────────────────┤
│ blocking │ 0.9259369s │ 1.1GB/s │
└──────────┴────────────┴────────────────┘