[
https://issues.apache.org/jira/browse/IGNITE-14743?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17383047#comment-17383047
]
Taras Ledkov commented on IGNITE-14743:
---------------------------------------
[~amashenkov], the patch is OK with me.
> Support Row with large values.
> ------------------------------
>
> Key: IGNITE-14743
> URL: https://issues.apache.org/jira/browse/IGNITE-14743
> Project: Ignite
> Issue Type: Improvement
> Reporter: Andrey Mashenkov
> Assignee: Andrey Mashenkov
> Priority: Major
> Labels: iep-54, ignite-3
> Fix For: 3.0.0-alpha3
>
> Attachments: Byte array columns only benchmark.txt, Fixlen cols only
> benchmark.txt, Latin1 string columns benchmark.txt, Non-latin string columns
> benchmark.txt, String marshalling comparison
>
> Original Estimate: 168h
> Time Spent: 2h 10m
> Remaining Estimate: 165h 50m
>
> h3. Motivation.
> For now, TupleAssembler writes offsets for varlen columns as 2-byte \{{short
> }}type.
> This implicitly restricts key/value sizes down to 64 kB in total.
> On another side, for small rows that columns can be addressed with \{{byte}}
> type, we will waste few bytes.
> h3. Description.
> Let's
> # allow 4 byte types (byte, short, int) for offsets.
> # implement and benchmark different approaches that allow us to write rows in
> the most compact way.
> # then choose and merge the best one.
> We can introduce several formats for writing Vartable (using byte/short/int
> offsets).
> Additional information about Vartable format can be coded into chunk flags.
> The first approach is to precalculate chunk total size, then choose the most
> compact format and write a chunk.
> The second approach is to write a chunk with the widest format then convert
> the chunk into the most compact format in place.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
