为什么批量插入内存优化的非持久表的速度与持久表相同？

Question

When running a large bulk insert to a durable and a non-durable memory optimized table, I'm getting the same speeds for both. Shouldn't the speed of of a bulk insert to a non-durable memory-optimized table be faster than a durable memory-optimized table? If so, what am I doing wrong here?

My test is as below, it takes ~30 seconds consistently. This is on SQL Server 2016 SP1. The bulk insert is 10 million rows from a csv file that I generated.

SQL

CREATE TABLE Users_ND (
  Id INT NOT NULL IDENTITY(1,1) PRIMARY KEY NONCLUSTERED HASH WITH (BUCKET_COUNT=10000000),  
  Username VARCHAR(200) NOT NULL  
) WITH (MEMORY_OPTIMIZED=ON, DURABILITY = SCHEMA_ONLY);  

CREATE TABLE Users_D (
  Id INT NOT NULL IDENTITY(1,1) PRIMARY KEY NONCLUSTERED HASH WITH (BUCKET_COUNT=10000000),  
  Username VARCHAR(200) NOT NULL  
) WITH (MEMORY_OPTIMIZED=ON, DURABILITY = SCHEMA_AND_DATA);  

SET STATISTICS TIME ON;
SET NOCOUNT ON;  

BULK INSERT Users_ND 
FROM 'users-huge.csv' 
WITH (FIRSTROW = 2, FIELDTERMINATOR = ',', ROWTERMINATOR = '\n', BATCHSIZE = 1000000);

BULK INSERT Users_D 
FROM 'users-huge.csv' 
WITH (FIRSTROW = 2, FIELDTERMINATOR = ',', ROWTERMINATOR = '\n', BATCHSIZE = 1000000);

users-huge.csv

Id, Username
,user1
,user2
...
,user10000000

Answer 1

事实证明，由于将大量插入的源文件保存在慢速的HDD上，所以出现了这个问题，因此读取文件时遇到了瓶颈。