New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Minimize memory internal fragmentation for Bloom filters #6427
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
@pdillinger has updated the pull request. Re-import the pull request |
1 similar comment
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It looks good! I am surprised how sophisticated it has become. Is there a way to simplify a little bit, e.g. randomized the choice, rather than looking at balance?
util/math.h
Outdated
} else { | ||
int lz = __builtin_clzll(static_cast<unsigned long long>(v)); | ||
return int{sizeof(unsigned long long)} * 8 - 1 - lz; | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Wow, this is so sophisticated.
Hmm. I don't know why I didn't give that much thought before. It should give similar aggregate behavior without the extra state tracking. |
Note to self: in order for CalculateSpace to predict the size returned by Finish (for partitioned filter), save a random threshold value in the builder object on construction and regenerate after each Finish. |
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
This change is probably of limited value unless we also change the block cache charge policy to charge for internal fragmentation, so that the block cache can more reliably use only its configured limit. (If people are already choosing setting expecting internal fragmentation overhead, they will probably need to re-tweak settings after such a change.) |
Summary: New experimental option BBTO::optimize_filters_for_memory builds filters that maximize their use of "usable size" from malloc_usable_size, which is also used to compute block cache charges. Rather than always "rounding up," we track state in the BloomFilterPolicy object to mix essentially "rounding down" and "rounding up" so that the average FP rate of all generated filters is the same as without the option. (YMMV as heavily accessed filters might be unluckily lower accuracy.) Thus, the option near-minimizes what the block cache considers as "memory used" for a given target Bloom filter false positive rate and Bloom filter implementation. There are no forward or backward compatibility issues with this change, though it only works on the format_version=5 Bloom filter. With Jemalloc, we see about 10% reduction in memory footprint (and block cache charge) for Bloom filters, but 1-2% increase in storage footprint, due to encoding efficiency losses (FP rate is non-linear with bits/key). Why not weighted random round up/down rather than state tracking? By only requiring malloc_usable_size, we don't actually know what the next larger and next smaller usable sizes for the allocator are. We pick a requested size, accept and use whatever usable size it has, and use the difference to inform our next choice. This allows us to narrow in on the right balance without tracking/predicting usable sizes. Why not weight history of generated filter false positive rates by number of keys? This could lead to excess skew in small filters after generating a large filter. Results with jemalloc (irrelevant details omitted): (normal keys/filter, but high variance) $ ./filter_bench -quick -impl=2 -average_keys_per_filter=30000 -vary_key_count_ratio=0.9 Build avg ns/key: 29.6278 Number of filters: 5516 Total size (MB): 200.046 Reported total allocated memory (MB): 220.597 Reported internal fragmentation: 10.2732% Bits/key stored: 10.0097 Average FP rate %: 0.965228 $ ./filter_bench -quick -impl=2 -average_keys_per_filter=30000 -vary_key_count_ratio=0.9 -optimize_filters_for_memory Build avg ns/key: 30.5104 Number of filters: 5464 Total size (MB): 200.015 Reported total allocated memory (MB): 200.322 Reported internal fragmentation: 0.153709% Bits/key stored: 10.1011 Average FP rate %: 0.966313 (very few keys / filter, optimization not as effective due to ~59 byte internal fragmentation in blocked Bloom filter representation) $ ./filter_bench -quick -impl=2 -average_keys_per_filter=1000 -vary_key_count_ratio=0.9 Build avg ns/key: 29.5649 Number of filters: 162950 Total size (MB): 200.001 Reported total allocated memory (MB): 224.624 Reported internal fragmentation: 12.3117% Bits/key stored: 10.2951 Average FP rate %: 0.821534 $ ./filter_bench -quick -impl=2 -average_keys_per_filter=1000 -vary_key_count_ratio=0.9 -optimize_filters_for_memory Build avg ns/key: 31.8057 Number of filters: 159849 Total size (MB): 200 Reported total allocated memory (MB): 208.846 Reported internal fragmentation: 4.42297% Bits/key stored: 10.4948 Average FP rate %: 0.811006 (high keys/filter) $ ./filter_bench -quick -impl=2 -average_keys_per_filter=1000000 -vary_key_count_ratio=0.9 Build avg ns/key: 29.7017 Number of filters: 164 Total size (MB): 200.352 Reported total allocated memory (MB): 221.5 Reported internal fragmentation: 10.5552% Bits/key stored: 10.0003 Average FP rate %: 0.969358 $ ./filter_bench -quick -impl=2 -average_keys_per_filter=1000000 -vary_key_count_ratio=0.9 -optimize_filters_for_memory Build avg ns/key: 30.7131 Number of filters: 160 Total size (MB): 200.928 Reported total allocated memory (MB): 200.938 Reported internal fragmentation: 0.00448054% Bits/key stored: 10.1852 Average FP rate %: 0.963387 Test Plan: unit test added, 'make check' with gcc, clang and valgrind
@pdillinger has updated the pull request. Re-import the pull request |
Simpler implementation that makes no assumptions other than using malloc_usable_size, as block cache does (previous comment inaccurate) |
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
usable_len = size_t{0xffffffc0}; | ||
} | ||
rv = (static_cast<uint32_t>(usable_len) & ~uint32_t{63}) + | ||
/* metadata */ 5; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Here is what man page https://man7.org/linux/man-pages/man3/malloc_usable_size.3.html says:
The value returned by malloc_usable_size() may be greater than the
requested size of the allocation because of alignment and minimum
size constraints. Although the excess bytes can be overwritten by
the application without ill effects, this is not good programming
practice: the number of excess bytes in an allocation depends on the
underlying implementation.
The main use of this function is for debugging and introspection.
Are you sure we want to use those bytes? At least consulting jemalloc friends before doing that.
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
include/rocksdb/table.h
Outdated
// Because some memory counted by block cache might be unmapped pages within | ||
// internal fragmentation, this option can increase observed RSS memory | ||
// usage. With cache_index_and_filter_blocks=true, this option makes the | ||
// block cache better at using space it is allowed. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can we mention in the comment that the implementation is against the best-practice suggestion in malloc_usable_size()'s man page?
@pdillinger has updated the pull request. Re-import the pull request |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@pdillinger has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
@pdillinger merged this pull request in 5b2bbac. |
Summary: New experimental option BBTO::optimize_filters_for_memory builds
filters that maximize their use of "usable size" from malloc_usable_size,
which is also used to compute block cache charges.
Rather than always "rounding up," we track state in the
BloomFilterPolicy object to mix essentially "rounding down" and
"rounding up" so that the average FP rate of all generated filters is
the same as without the option. (YMMV as heavily accessed filters might
be unluckily lower accuracy.)
Thus, the option near-minimizes what the block cache considers as
"memory used" for a given target Bloom filter false positive rate and
Bloom filter implementation. There are no forward or backward
compatibility issues with this change, though it only works on the
format_version=5 Bloom filter.
With Jemalloc, we see about 10% reduction in memory footprint (and block
cache charge) for Bloom filters, but 1-2% increase in storage footprint,
due to encoding efficiency losses (FP rate is non-linear with bits/key).
Why not weighted random round up/down rather than state tracking? By
only requiring malloc_usable_size, we don't actually know what the next
larger and next smaller usable sizes for the allocator are. We pick a
requested size, accept and use whatever usable size it has, and use the
difference to inform our next choice. This allows us to narrow in on the
right balance without tracking/predicting usable sizes.
Why not weight history of generated filter false positive rates by
number of keys? This could lead to excess skew in small filters after
generating a large filter.
Results from filter_bench with jemalloc (irrelevant details omitted):
And from db_bench (block cache) with jemalloc:
(Due to specific key density, this example tends to generate filters that are "worse than average" for internal fragmentation. "Better than average" cases can show little or no improvement.)
Test Plan: unit test added, 'make check' with gcc, clang and valgrind