On Sat, Nov 11, 2017 at 3:25 AM, Robert Haas wrote:
> On Thu, Nov 9, 2017 at 3:55 AM, amul sul wrote:
>> It took me a little while to understand this calculation. You have moved
>> this
>> code from tbm_create(), but I think you should move the following
>> comment as well:
>
> I made an adjust
On Thu, Nov 9, 2017 at 3:55 AM, amul sul wrote:
> It took me a little while to understand this calculation. You have moved this
> code from tbm_create(), but I think you should move the following
> comment as well:
I made an adjustment that I hope will address your concern here, made
a few other
Hi Dilip,
v6 patch:
42 + /*
43 +* Estimate number of hashtable entries we can have within
maxbytes. This
44 +* estimates the hash cost as sizeof(PagetableEntry).
45 +*/
46 + nbuckets = maxbytes /
47 + (sizeof(PagetableEntry) + sizeof(Pointer) + sizeof(Pointer));
It too
On Fri, Oct 6, 2017 at 9:21 PM, Dilip Kumar wrote:
> On Fri, Oct 6, 2017 at 7:24 PM, Dilip Kumar wrote:
>> On Fri, Oct 6, 2017 at 6:08 PM, Alexander Kuzmenkov
>> wrote:
>>>
Analysis: The estimated value of the lossy_pages is way higher than
its actual value and reason is that the total
On Fri, Oct 6, 2017 at 7:24 PM, Dilip Kumar wrote:
> On Fri, Oct 6, 2017 at 6:08 PM, Alexander Kuzmenkov
> wrote:
>>
>>> Analysis: The estimated value of the lossy_pages is way higher than
>>> its actual value and reason is that the total_pages calculated by the
>>> "Mackert and Lohman formula" i
On Fri, Oct 6, 2017 at 7:04 PM, Dilip Kumar wrote:
> On Fri, Oct 6, 2017 at 6:36 PM, Robert Haas wrote:
>> On Fri, Oct 6, 2017 at 2:12 AM, Dilip Kumar wrote:
The performance results look good, but that's a slightly different
thing from whether the estimate is accurate.
+n
On Fri, Oct 6, 2017 at 6:08 PM, Alexander Kuzmenkov
wrote:
>
>> Analysis: The estimated value of the lossy_pages is way higher than
>> its actual value and reason is that the total_pages calculated by the
>> "Mackert and Lohman formula" is not correct.
>
>
> I think the problem might be that the t
On Fri, Oct 6, 2017 at 6:36 PM, Robert Haas wrote:
> On Fri, Oct 6, 2017 at 2:12 AM, Dilip Kumar wrote:
>>> The performance results look good, but that's a slightly different
>>> thing from whether the estimate is accurate.
>>>
>>> +nbuckets = tbm_calculate_entires(maxbytes);
>>>
>>> entires?
On Fri, Oct 6, 2017 at 2:12 AM, Dilip Kumar wrote:
>> The performance results look good, but that's a slightly different
>> thing from whether the estimate is accurate.
>>
>> +nbuckets = tbm_calculate_entires(maxbytes);
>>
>> entires?
>
> changed to
> + tbm->maxentries = (int) tbm_calculate_en
Analysis: The estimated value of the lossy_pages is way higher than
its actual value and reason is that the total_pages calculated by the
"Mackert and Lohman formula" is not correct.
I think the problem might be that the total_pages includes cache effects
and rescans. For bitmap entries we sh
On Thu, Oct 5, 2017 at 8:15 PM, Robert Haas wrote:
> On Sun, Sep 17, 2017 at 7:04 AM, Dilip Kumar wrote:
>> I used lossy_pages = max(0, total_pages - maxentries / 2). as
>> suggesed by Alexander.
>
> Does that formula accurately estimate the number of lossy pages?
I have printed the total_pages
On Sun, Sep 17, 2017 at 7:04 AM, Dilip Kumar wrote:
> I used lossy_pages = max(0, total_pages - maxentries / 2). as
> suggesed by Alexander.
Does that formula accurately estimate the number of lossy pages?
The performance results look good, but that's a slightly different
thing from whether the
On Sun, Sep 17, 2017 at 4:34 PM, Dilip Kumar wrote:
>>
>> I have repeated one of the tests after fixing the problems pointed by
>> you but this time results are not that impressive. Seems like below
>> check was the problem in the previous patch
>>
>>if (tbm->nentries > tbm->maxentries / 2)
>
On Mon, Sep 4, 2017 at 11:18 AM, Dilip Kumar wrote:
> On Thu, Aug 31, 2017 at 11:27 PM, Robert Haas wrote:
>
> I have repeated one of the tests after fixing the problems pointed by
> you but this time results are not that impressive. Seems like below
> check was the problem in the previous patch
On Thu, Aug 31, 2017 at 11:27 PM, Robert Haas wrote:
I have repeated one of the tests after fixing the problems pointed by
you but this time results are not that impressive. Seems like below
check was the problem in the previous patch
if (tbm->nentries > tbm->maxentries / 2)
tbm->max
On Thu, Aug 31, 2017 at 2:26 AM, Dilip Kumar wrote:
> Thanks for the feedback. I will work on it.
Another thought is that you probably want/need to test across a range
of work_mem values.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
--
Sent via
On Wed, Aug 30, 2017 at 2:00 AM, Robert Haas wrote:
>
> I suggest defining a TBM_FILLFACTOR constant instead of repeating the
> value 0.9 in a bunch of places. I think it would also be good to try
> to find the sweet spot for that constant. Making it bigger reduces
> the number of lossy entries
On Tue, Aug 29, 2017 at 1:08 AM, Dilip Kumar wrote:
> (Time in ms)
> Queryhead patch
>
> 6 2381914571
> 14 1351411183
> 15 49980 32400
> 20204441 188978
These are cool results, but this patch is obviously not ready for
p
On Wed, Aug 23, 2017 at 9:45 AM, Dilip Kumar wrote:
>>
>> ...
>> if (tbm->nentries <= tbm->maxentries / 2)
>> {
>> /*
>> * We have made enough room.
>> ...
>> I think we could try higher fill factor, say, 0.9. tbm_lossify basically
>> just
On Tue, Aug 22, 2017 at 8:40 PM, Alexander Kumenkov
wrote:
> Hi Dilip,
>
> Recently I was thinking about this too, when working on the index-only
> count(*) patch for indexes supporting amgetbitmap [1]. That patch teaches
> bitmap heap scan node to skip heap fetches under certain conditions. Exact
Hi Dilip,
Recently I was thinking about this too, when working on the index-only
count(*) patch for indexes supporting amgetbitmap [1]. That patch
teaches bitmap heap scan node to skip heap fetches under certain
conditions. Exact tidbitmap pages are a prerequisite for this, so the
lossines of
On Thu, Aug 17, 2017 at 12:06 AM, Dilip Kumar wrote:
> I have attempted a very simple POC with this approach just to see how
> many lossy pages we can save if we lossify all the pages falling in
> the same chunk first, before moving to the next page. I have taken
> some data on TPCH scale 20 with
On Wed, Jul 26, 2017 at 10:35 PM, Robert Haas wrote:
> On Thu, Jun 8, 2017 at 10:44 AM, Dilip Kumar wrote:
>> On Thu, May 18, 2017 at 8:07 PM, Robert Haas wrote:
>>
Thanks for the feedback. I haven't yet worked on optimizer side
feedback but I have done some work for improving the executor par
On Thu, Jun 8, 2017 at 10:44 AM, Dilip Kumar wrote:
> On Thu, May 18, 2017 at 8:07 PM, Robert Haas wrote:
>
> Thanks for the feedback and sorry for the delayed response.
>
>> You might need to adjust effective_cache_size.
>
> You are right. But, effective_cache_size will have the impact on the
>
On Thu, May 18, 2017 at 8:07 PM, Robert Haas wrote:
Thanks for the feedback and sorry for the delayed response.
> You might need to adjust effective_cache_size.
You are right. But, effective_cache_size will have the impact on the
number of pages_fetched when it's used as parameterized path (i.e
On Thu, May 18, 2017 at 2:52 AM, Dilip Kumar wrote:
> Most of the queries show decent improvement, however, Q14 shows
> regression at work_mem = 4MB. On analysing this case, I found that
> number of pages_fetched calculated by "Mackert and Lohman formula" is
> very high (1112817) compared to the a
I would like to propose a patch to improve the cost of bitmap heap
scan that is sensitive to work_mem. Currently, in bitmap scan, we
don't consider work_mem. Now, in cases when there are a lot of lossy
pages bitmap scan gets selected that eventually leads to degraded
performance.
While evaluating
27 matches
Mail list logo