On Thu, Nov 09, 2017 at 12:23:24PM +0100, Michal Židek wrote:
> On 11/08/2017 11:28 AM, Sumit Bose wrote:
> > Hi,
> > 
> > while trying to prepare some diagrams to illustrate how the memory mapped
> > cache works I realized that we might not use the payload factor as
> > expected.
> > 
> > In sss_mmap_cache_init() we set a payload depending of the type of
> > cached data, e.g. for passwd data 4 * MC_SLOT_SIZE or for group data 3 *
> > MC_SLOT_SIZE. This value is used to calculate the data table size:
> > 
> > #define MC_DT_SIZE(elems, payload) ( (elems) * (payload) )
> > 
> >      mc_ctx->dt_size = MC_DT_SIZE(n_elem, payload);
> > 
> > since this does not change the slot size there is now room for e.g
> > (n_elem * 4) slots for passwd data, which makes sense since we assume
> > that a typical passwd entry will occupy 4 slot and we want the cache to
> > be able to store n_elem passwd entries.
> > 
> > But the memory for the tree table is calculated only with n_elem
> > 
> > #define MC_FT_SIZE(elems) ( (elems) / 8 )
> > 
> >      mc_ctx->ft_size = MC_FT_SIZE(n_elem);
> > 
> > This means that although we allocated memory for e.g. (4 * n_elem) slots
> > we can only handle n_elem slots because there are only n_elem bits in
> > the free table. As a result 3/4 of the allocate memory for the data is
> > unused and entries in the cache are overwritten earlier than needed.
> > 
> > Do you agree with this or did I miss something in the remaining part of
> > the code which makes sure the memory allocated for the data table can be
> > used completely?
> > 
> > If not I would suggest to make the payload only a factor which then can
> > be used to allocate the memory for the data table and the free table.
> > 
> > bye,
> > Sumit
> 
> Hi Sumit,
> 
> I looked into it and I think you are correct. Will you file
> a ticket?

yes, I just opened https://pagure.io/SSSD/sssd/issue/3572

Thank you for checking this as well.

bye,
Sumit

> 
> Michal
> _______________________________________________
> sssd-devel mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
_______________________________________________
sssd-devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to