#benchmark

Overview

This directory contains Ledger trace-based benchmarks. Each benchmark is implemented as a client connected to Ledger and using its API to perform one or more of its basic operations (creating a page, writing/deleting entries, ...).

Building benchmarks

To build the benchmarks, add //src/ledger/bin/tests/benchmark to the list of packages available in your build:

fx set core.x64 --release --with //src/ledger/bin/tests/benchmark
fx build ledger_benchmarks

Run benchmarks

Using trace spec files

The easiest way to run a benchmark is using the associated trace spec file, for example:

trace record --spec-file=/pkgfs/packages/ledger_benchmarks/0/data/put.tspec

You can read more on the trace spec file format in the trace-based benchmarking guide.

Tracing benchmark apps directly

You can also trace the app directly:

trace record --categories=benchmark,ledger \
  fuchsia-pkg://fuchsia.com/ledger_benchmarks#meta/put.cmx \
  --entry-count=10 --transaction-size=1 --key-size=100 --value-size=100 \
  --refs=false

That would generate a trace result file (default: /data/trace.json), which can be analysed manually or using special tools.

Benchmark parameters

*.tspec files specify the parameters with which benchmark apps are run. You can override these by passing the “--append-args” argument to the trace record tool. Commonly used among all the ledger benchmarks are the following parameters:

  • entry-count for the number of entries to perform operations on
  • unique-key-count if the number of operations and the number of entries differ (i.e. some entries are overwritten), this denotes the number of unique entries in the page
  • key-size for the size of the key (in bytes)
  • value-size for the size of the value (in bytes)
  • refs with the values on or off for the reference strategy: if set to on, entries will be put using CreateReference/PutAsReference, otherwise they will be treated as inline entries.
  • commit-count for the number of commits made to the page
  • transaction-size for the number of operations in a single transaction

Unless the name of the benchmark suggest otherwise, default values are:

  • 100 entries
  • key size 100
  • value size 1000 Benchmarks under sync and convergence use smaller number of entries and smaller value size.

A note regarding benchmark apps

Since the benchmark apps are designed to be used with tracing, running them without a tracing will not generate any results.

List of benchmarks

Some of the benchmark apps have several corresponding tspec files, to exercise different possible scenarios of using Ledger. For example, get_page benchmark is used in add_new_page.tspec to emulate creating many pages with different IDs, and in get_same_page.tspec to create several connections to the same page.

Local benchmarks

  • Get page: How long does it take to establish a new page connection?
    • add_new_page.tspec: connection to the new page with previously unused ID.
    • add_new_page_precached.tspec: same as above, but waits for a precached Ledger Page to be ready before each request.
    • add_new_page_after_clear.tspec: same as add_new_page.tspec but each previous page is populated with one entry, then cleared and evicted.
    • get_same_page.tspec: several connections to the same page
    • get_page_id.tspec: how long does the GetId() call takes on a newly created page?
  • Put: How long does it take to write data to a page? And how long before the client will receive a PageWatcher notification about its own change?
    • put.tspec: basic case
    • put_big_entry.tspec: writing big entries to the page
    • put_as_reference.tspec: entries are put as references (CreateReference + PutReference)
    • transaction.tspec: changes are made in a transaction
  • Update entry: How long does it take to update an existing value in Ledger (make several Put operations with the same key, but different values)?
    • update_entry.tspec: basic case
    • update_big_entry.tspec: put an entry of big size, then update its value
    • update_entry_transactions.tspec: changes are grouped in transactions
  • Get entry: How long does it take to retrieve a single value from a page? This benchmark also measures the time taken by the GetSnapshot call and the GetKeys call to retrieve all keys from the page.
    • get_small_entry.tspec: basic case
    • get_small_entry_inline.tspec: GetInline is used instead of Get
    • get_big_entry.tspec: values of bigger size are used.
  • Delete entry: How long does it take to delete an entry from a page?
    • delete_entry.tspec: each entry is deleted separately (outside of a transaction)
    • delete_big_entry.tspec: same as above, but for big entries
    • delete_entry_transactions.tspec: deletions are grouped in transactions
    • disk_space_cleared_page.tspec: how much space does ledger take after the page was cleared out (all the entries deleted in one transaction)?
  • Story simulation: How long does it take to create and edit a Modular story?
    • stories_single_active.tspec: a single story is active at all times. Once editing is complete, the contents of the story (and any corresponding rows in the root page) are removed. This is the base case.
    • stories_many_active.tspec: same as in the base case, but keep up to 20 pages active, before clearing their contents.
    • stories_wait_cached.tspec: same as in the base case, but wait for a precached Ledger Page before creating a new Story.
  • Disk space: How much disk space does ledger use to store pages, objects and commits?
    • disk_space_empty_ledger.tspec: empty ledger (with no pages)
    • disk_space_empty_pages.tspec: ledger containing only empty pages
    • disk_space_entries.tspec: ledger with one page containing some entries
    • disk_space_small_keys.tspec: same, but with small (10 bytes) keys.
    • disk_space_updates.tspec: ledger with one page containing only one entry, but long commit history
    • disk_space_one_commit_per_entry.tspec: ledger with one page containing several entries, each of them added in a separate commit