Thanks Matthew,

What I need is "bulk processing" of documents. I currently run LO in headless 
mode and open the documents in read-only format (macros disabled) hoping that 
would improve performance. To answer to your points: no interactiveness and 
high throughput to scrape document content.

Your suggestion of running a pool of LO instances seems to be implemented by 
the oodaemon (https://sourceforge.net/projects/oodaemon/) that I mentioned 
earlier, but the project seems quite dated.

Cheers,
Jens


> On Oct 20, 2017, at 13:24, Matthew J. Francis <[email protected]> wrote:
> 
> As I mentioned before, while a single LO process can serve multiple clients 
> simultaneously, and with low latency for starting a request, it will likely 
> struggle to fully utilise all the threads of a modern CPU.
> 
> I believe you still haven't mentioned what the specific use case you're 
> trying to achieve is - what are its requirements for interactiveness and/or 
> throughput? (i.e. is it an interactive process that will primarily require 
> low latency, a bulk process that requires high overall throughput but may not 
> mind if individual requests are delayed for a short time, or something that 
> requires both low request latency and high throughput?)
> 
> If you need both the lowest request latency and the highest throughput on a 
> single machine, you may need to try adopting a hybrid model - start a pool of 
> independent LO instances of an appropriate size for your number of CPU 
> threads, then feed them jobs one at a time each. There is definitely no 
> canned answer for this, so you would have to do your own benchmarking and 
> tuning to find the optimum pool size for the resources of your local machine.
> 
> Regards
> Matthew Francis

--
Jens Tröger
http://savage.light-speed.de/

_______________________________________________
LibreOffice mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/libreoffice

Reply via email to