cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sylvain Lebresne (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-2843) better performance on long row read
Date Wed, 20 Jul 2011 08:08:59 GMT


Sylvain Lebresne commented on CASSANDRA-2843:

bq. Doesn't it make sense then to change the AL fallback-to-bsearch into an assertion failure?

I kind of have a small preference for not turning it into a failure, but if we don't see a
use case where we would want to use AL even though not all input are sorted in the near future,
I could be convinced to doing it for now. There is one such use case though that I think could
be reasonable, and that's the new SSTableSimple*Writer classes introduced by CASSANDRA-2911.
The thing is, you need to have quite a bit of input out of order for AL to start being slower
than the alternatives (even a TreeMap one). I tried a modified version of my micro-benchmark
adding columns in reverse sorted order (worst case for AL), and on my machine, you need to
add 1000 columns to start seeing AL slower than the others. Which leads me to believe that
AL could be a good default for the CF used in SSTableSimple*Writer.

bq. I see a typo in Sylvain's patch, and added a comment on ArrayBackedColumnMap: its addAll()
does assume sorted input

AddAll does assume sorted input, but that's because IColumnMap is meant to be a sorted map
(IColumnMap doesn't extend SortedMap for technical reason, but it is really meant as a SortedMap).
That is, I agree that comments could be improved, but I think that it is the IColumnMap one
that needs to be fixed by specifying that the iterator should iter in sorted order. Anyway,
that cosmetic and could be changed during commit if not before.

bq. AbstractColumnContainer.DeletionInfo has to be protected, otherwise eclipse gives a compile

Is there another reason than "otherwise eclipse gives a compile error". Because it doesn't
seem very related to this patch and I'm keen on not changing stuff randomly without understanding

bq.  in the 95th percentile there was an approximate 10% gain across the board

Great, thanks Brandon for the testing.

> better performance on long row read
> -----------------------------------
>                 Key: CASSANDRA-2843
>                 URL:
>             Project: Cassandra
>          Issue Type: New Feature
>            Reporter: Yang Yang
>         Attachments: 2843.patch, 2843_c.patch, fast_cf_081_trunk.diff, incremental.diff,
> currently if a row contains > 1000 columns, the run time becomes considerably slow
(my test of 
> a row with 30 00 columns (standard, regular) each with 8 bytes in name, and 40 bytes
in value, is about 16ms.
> this is all running in memory, no disk read is involved.
> through debugging we can find
> most of this time is spent on 
> [Wall Time]  org.apache.cassandra.db.Table.getRow(QueryFilter)
> [Wall Time]  org.apache.cassandra.db.ColumnFamilyStore.getColumnFamily(QueryFilter, ColumnFamily)
> [Wall Time]  org.apache.cassandra.db.ColumnFamilyStore.getColumnFamily(QueryFilter, int,
> [Wall Time]  org.apache.cassandra.db.ColumnFamilyStore.getTopLevelColumns(QueryFilter,
int, ColumnFamily)
> [Wall Time]  org.apache.cassandra.db.filter.QueryFilter.collectCollatedColumns(ColumnFamily,
Iterator, int)
> [Wall Time]  org.apache.cassandra.db.filter.SliceQueryFilter.collectReducedColumns(IColumnContainer,
Iterator, int)
> [Wall Time]  org.apache.cassandra.db.ColumnFamily.addColumn(IColumn)
> ColumnFamily.addColumn() is slow because it inserts into an internal concurrentSkipListMap()
that maps column names to values.
> this structure is slow for two reasons: it needs to do synchronization; it needs to maintain
a more complex structure of map.
> but if we look at the whole read path, thrift already defines the read output to be List<ColumnOrSuperColumn>
so it does not make sense to use a luxury map data structure in the interium and finally convert
it to a list. on the synchronization side, since the return CF is never going to be shared/modified
by other threads, we know the access is always single thread, so no synchronization is needed.
> but these 2 features are indeed needed for ColumnFamily in other cases, particularly
write. so we can provide a different ColumnFamily to CFS.getTopLevelColumnFamily(), so getTopLevelColumnFamily
no longer always creates the standard ColumnFamily, but take a provided returnCF, whose cost
is much cheaper.
> the provided patch is for demonstration now, will work further once we agree on the general
> CFS, ColumnFamily, and Table  are changed; a new FastColumnFamily is provided. the main
work is to let the FastColumnFamily use an array  for internal storage. at first I used binary
search to insert new columns in addColumn(), but later I found that even this is not necessary,
since all calling scenarios of ColumnFamily.addColumn() has an invariant that the inserted
columns come in sorted order (I still have an issue to resolve descending or ascending  now,
but ascending works). so the current logic is simply to compare the new column against the
end column in the array, if names not equal, append, if equal, reconcile.
> slight temporary hacks are made on getTopLevelColumnFamily so we have 2 flavors of the
method, one accepting a returnCF. but we could definitely think about what is the better way
to provide this returnCF.
> this patch compiles fine, no tests are provided yet. but I tested it in my application,
and the performance improvement is dramatic: it offers about 50% reduction in read time in
the 3000-column case.
> thanks
> Yang

This message is automatically generated by JIRA.
For more information on JIRA, see:


View raw message