[
https://issues.apache.org/jira/browse/PDFBOX-3088?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14992434#comment-14992434
]
Tilman Hausherr commented on PDFBOX-3088:
-----------------------------------------
I'm currently doing some statistics with my test set to find out what / how
much is cached. The solution will likely be:
- don't cache for fonts with huge glyph tables
- (maybe) don't cache certain type of fonts (e.g. subsets?)
- (maybe) don't cache more than NNN glyphs
Some sizes:
{quote}
numGlyphs: 3418, font: TimesNewRomanPS-BoldMT
numGlyphs: 3419, font: ArialMT
numGlyphs: 20458, font: MS-Gothic
numGlyphs: 52487, font: DroidSansFallback
numGlyphs: 34046, font: MingLiU
numGlyphs: 34046, font: PMingLiU
{quote}
We'd like to cache Arial and Times, obviously. We don't need to cache
MS-Gothic, which is just a fallback for Zapf Dingbats.
> Cache glyph table to optimize concurrent access
> -----------------------------------------------
>
> Key: PDFBOX-3088
> URL: https://issues.apache.org/jira/browse/PDFBOX-3088
> Project: PDFBox
> Issue Type: Improvement
> Components: FontBox
> Affects Versions: 2.0.0
> Reporter: ccouturi
> Assignee: Tilman Hausherr
> Priority: Minor
> Labels: Optimization
> Fix For: 2.0.0
>
> Attachments: 0001-PDFBOX-3088-cache-glyph-table.patch,
> Benchmark.java, BenchmarkPDFBox3088.java, test_medium.pdf
>
>
> If several threads convert several pdf to png (one thread access to a single
> document at a time) they are a contention on a lock in GlythTable. Jstack
> shows that all threads are in state blocked on the synchronized block in the
> getGlyph method. The lock is necessary, it's ok, but degrades performance.
> This patch cache glyphs already read.
> With the patch PDFBOX-3080, the follow benchmark compare 1000 pdf conversions
> with 1, 8, and 50 threads.
> || Simulation|| PDF 2.0-SNAPSHOT || With this patch + PDFBOX3080 ||
> || 1000 conversions / 1 thread | 120 s | 71 s|
> || 1000 conversions / 8 threads | 76 s | 28 s|
> || 1000 conversions / 50 threads | 81 s | 33 s|
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
