[Opensim-dev] Memory cache
diva at metaverseink.com
diva at metaverseink.com
Thu May 21 22:34:01 UTC 2009
Mic Bowman wrote:
> Maybe I'm missing something because I have no idea what I'm "right" about...
Hehehe. "You're right" in pointing to that piece of code, which is not
used anymore, and it's critically important for this puzzle :-) That old
code is Framework.Cache.AssetServerBase and Framework.Cache.AssetCache.
The latter is actually not the cache (errr...), the real cache itself is
Framework.Cache.
> As I understand it... somewhere along the way 1) the old cache was
> removed and 2) the new cache was never turned on correctly.
The real cache is exactly the same as before, Framework.Cache. I'm not
even going to try to explain the old names AssetServerBase (foo) and
AssetCache (bar), let's just say that foo and bar are not being used
anymore. Instead, there is an IAssetService in place that deals with
getting/posting assets from/to the asset service, and may or may not
have a memory cache in place -- it's optional now.
(If you want to look at its implementation, it's in
OpenSim.Services.Connectors.AssetServicesConnector)
> Regarding the performance observation... Having looked at what appears
> to be the old version of the cache... I can easily understand why the
> simulator performance got better when it was removed. It was... how to
> be diplomatic... an unusual implementation... and subject to long
> delays (it looks as though with a single client you would get about 1
> asset per 500 ms on avg). I don't see any evidence that the new cache
> implementation actually improves performance... just that turning off
> the old one does.
See my clarification above regarding the... how should I put it...
unusual names that were being used. But yes, your diplomatic observation
is right on. And now things are done in a less unusual manner that
speeds asset fetching probably by a factor of 10 (guestimate).
> Regarding the consistency in asset server signature... This is
> disturbing. If the changing the asset cache does not affect the rate
> of requests on the asset server then one of three situations is
> occurring... [...]
Someone needs to do some digging and testing about this...
More information about the Opensim-dev
mailing list