Hi Tristan,
On 16/05/13 15:28, Tristan Roddis wrote:
We are planning to set up a new machine to run Jena TDB with Fuseki, but
the hosting organisation has a strong preference for Windows instead of
Linux.
When I spoke to Andy Seaborne about this at the recent W3C Open Data
conference, he mentioned that Windows has issues with the use of memory
mapped files, and I just wanted to check what the implications of this are.
I saw a couple of references to this in the list archives, e.g.:
http://mail-archives.apache.org/mod_mbox/jena-users/201103.mbox/%3cof213233fa.dfef8523-on8525785e.006277bd-8525785e.00652...@ca.ibm.com%3E
Caution: Mar 2011
However, this didn't really clarify things for me!
What I'd really like to know is:
- is it really that much worse to run Jena on Windows than Linux?
Not "much worse" - just not as good.
The key issues are:
1/ Memory mapped files don't seem to be faster than doing the caching by
copying (comparing a special version of TDB with enlarged in-heap caches
to get a true comparison).
2/ You can't delete a database, even if closed and expelled from the
storage manager, without the JVM exiting. This is a well-know and quite
old Java issue. There are some workarounds that could be applied but
the best are sun-JVM specific. I haven't tried any in real systems.
Or, more specifically, any answers to questions like:
- if yes, how much worse[1]?
Slower, not unstable.
- should any Windows OS (Win7, Win2008 server) or variant (64 or 32 bit)
be favoured over another?
Hopefully, someone has figures (we run Linux in production for all sorts
of reasons).
It's still better in Mar 2011general to use 64bit mode unless you have
many small databases Mar 2011and/or delete databases while running.
- any machine specs that should be favoured (e.g. RAM 3x the TDB index
size)?>
Not special to windows - RAM about the size of the TDB indexes (at least
2 of them) and whatever for the JVM heap. Don't make the heap very large.
- any configuration tweaks that should be applied?
- best ways to benchmark performance and ensure it all works on Windows?
As ever, benchmarks on your usage patterns are the only real benchmarks.
Thanks in advance,
-Tristan.
[1] e.g. is the issue just one of consistently slower performance (which
we could probably handle, so long as we know how slow!) or one of
instability (which I would rather avoid)?
Andy
