Re: POC: Cache data in GetSnapshotData()

On Wed, Sep 13, 2017 at 7:24 PM, Jesper Pedersen
<jesper(dot)pedersen(at)redhat(dot)com> wrote:
> I have done a run with this patch on a 2S/28C/56T/256Gb w/ 2 x RAID10 SSD
> machine.
>
> Both for -M prepared, and -M prepared -S I'm not seeing any improvements (1
> to 375 clients); e.g. +-1%.

My test was done on an 8 socket NUMA intel machine, where I could
clearly see improvements as posted before.
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 128
On-line CPU(s) list: 0-127
Thread(s) per core: 2
Core(s) per socket: 8
Socket(s): 8
NUMA node(s): 8
Vendor ID: GenuineIntel
CPU family: 6
Model: 47
Model name: Intel(R) Xeon(R) CPU E7- 8830 @ 2.13GHz
Stepping: 2
CPU MHz: 1064.000
BogoMIPS: 4266.62
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 24576K
NUMA node0 CPU(s): 0,65-71,96-103
NUMA node1 CPU(s): 72-79,104-111
NUMA node2 CPU(s): 80-87,112-119
NUMA node3 CPU(s): 88-95,120-127
NUMA node4 CPU(s): 1-8,33-40
NUMA node5 CPU(s): 9-16,41-48
NUMA node6 CPU(s): 17-24,49-56
NUMA node7 CPU(s): 25-32,57-64

Let me recheck if the improvement is due to NUMA or cache sizes.
Currently above machine is not available for me. It will take 2 more
days for same.

> Although the -M prepared -S case should improve, I'm not sure that the extra
> overhead in the -M prepared case is worth the added code complexity.

Each tpcb like (-M prepared) transaction in pgbench have 3 updates 1
insert and 1 select statements. There will be more GetSnapshotData
calls than the end of the transaction (cached snapshot invalidation).
So I think whatever we cache has a higher chance of getting used
before it is invalidated in -M prepared. But on worst cases where we
have quick write transaction which invalidates cached snapshot before
it is reused becomes an overhead.
As Andres has suggested previously I need a mechanism to identify and
avoid caching for such scenarios. I do not have a right solution for
this at present but one thing we can try is just before caching if we
see an exclusive request in wait queue of ProcArrayLock we can avoid
caching.

--
Thanks and Regards
Mithun C Y
EnterpriseDB: http://www.enterprisedb.com