Re: Relation extension scalability

On Mon, Feb 29, 2016 at 3:37 PM, Dilip Kumar <dilipbalaut(at)gmail(dot)com> wrote:

>
> On Wed, Feb 10, 2016 at 7:06 PM, Dilip Kumar <dilipbalaut(at)gmail(dot)com>
> wrote:
>
>>
>
> Test2: Identify that improvement in case of multiextend is becuase of
> avoiding context switch or some other factor, like reusing blocks b/w
> backend by putting in FSM..
>
> 1. Test by just extending multiple blocks and reuse in it's own backend
> (Don't put in FSM)
> Insert 1K record data don't fits in shared buffer 512MB Shared Buffer
>
>
>
Client Base Extend 800 block self use Extend 1000 Block
> 1 117 131 118
> 2 111 203 140
> 3 51 242 178
> 4 51 231 190
> 5 52 259 224
> 6 51 263 243
> 7 43 253 254
> 8 43 240 254
> 16 40 190 243
>
> We can see the same improvement in case of self using the blocks also, It
> shows that Sharing the blocks between the backend was not the WIN but
> avoiding context switch was the measure win.
>
>
One thing that is slightly unclear is that whether there is any overhead
due to buffer eviction especially when the buffer to be evicted is already
dirty and needs XLogFlush(). One reason why it might not hurt is that by
the time we tries to evict the buffer, corresponding WAL is already flushed
by WALWriter or other possibility could be that even if it is getting done
during buffer eviction, the impact for same is much lesser. Can we try to
measure the number of flush calls which happen during buffer eviction?

> 2. Tested the Number of ProcSleep during the Run.
> This is the simple script of copy 10000 record in one transaction of size
> 4 Bytes
>
> * BASE CODE*
> *PATCH MULTI EXTEND*
> Client Base_TPS ProcSleep Count Extend By 10 Block Proc
> Sleep Count
> 2 280 457,506
> 311 62,641
> 3 235 1,098,701
> 358 141,624
> 4 216 1,155,735
> 368 188,173
>
> What we can see in above test that, in Base code performance is degrading
> after 2 threads, while Proc Sleep count in increasing with huge amount.
>
> Compared to that in Patch, with extending 10 blocks at a time Proc Sleep
> reduce to ~1/8 and we can see it is constantly scaling.
>
> Proc Sleep test for Insert test when data don't fits in shared buffer and
> inserting big record of 1024 bytes, is currently running
> once I get the data will post the same.
>
>
Okay. However, I wonder if the performance data for the case when data
doesn't fit into shared buffer also shows similar trend, then it might be
worth to try by doing extend w.r.t load in system. I mean to say we can
batch the extension requests (as we have done in ProcArrayGroupClearXid)
and extend accordingly, if that works out then the benefit could be that we
don't need any configurable knob for the same.

With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com