Re: Impact of persistent NVM Memory (pnvm) on ZODB's performance

Sean Upton <[email protected]> Fri, 6 Oct 2017 13:36:33 -0600
Newsgroups gmane.comp.web.zope.zodb
Message-ID <[email protected]>
I have not used such devices, but from my limited investigation, I believe that such device APIs (let's say, pynvm [1] and NVML) require a different programming model that will be:

(1) Hard work; you have to do a fair bit of programming using APIs that may or may not match higher-level assumptions about your storage model.

(2) Don't get you out of considering serialization.

(3) Have complications around concurrency and consistency that are thorny [2].

This is likely a case where the investment may be onerous and provide limited returns vs aggressive caching in DRAM.   The likely parties that could afford such an R&D tradeoff are the device manufacturers themselves, I suppose.


[1] https://github.com/perone/pynvm

[2] http://www.hpl.hp.com/techreports/2015/HPL-2015-59.pdf


Sean

> On Oct 6, 2017, at 12:32 PM, Yashwardhan Singh <[email protected]> wrote:
> 
> I was thinking about non-volatile memory as in nvram (Non-volatile RAM). Optane is High speed persistent storage (SSD), where as  NVRAM sits on memory bus along with DDR RAM. So object created on that memory will be much faster to access than SSDs.
>     http://pmem.io/ 
> 
> -- 
> You received this message because you are subscribed to the Google Groups "zodb" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to [email protected] <mailto:[email protected]>.
> For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.

-- 
You received this message because you are subscribed to the Google Groups "zodb" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
For more options, visit https://groups.google.com/d/optout.