I want to know whether there is any operations before SELECT. When I CREATE a new table and COPY some data to this table, it will spend many seconds to SELECT in first time, and when I excute it the next time, it will spend less time. What’s more,my GPU is 4 * Tesla P100. It will spend 1000+ ms to excute the fellow SELECT in 1.2B database(The database is what you used in http://stat-computing.org/dataexpo/2009/the-data.html):
select UniqueCarrier,TailNum,avg(ArrDelay),max(fMonth) from flights_ASA_year group by UniqueCarrier,TailNum order by max(fMonth);
I think it is slower than your Official data. So is there any operations i have forgeted. For example, insert the data into GPU memury before excute select.
When you COPY data into a table, this data is written to disk and not automatically loaded into GPU memory until a query requires it for processing. Time for that load depends on quantity of data needed and bandwidth, so you will find this is faster with SSD than spinning disks, and faster with NVLINK than PCIE.
Thanks for exploring MapD.
You may also wan to read this thread as it goes into more detail regarding he initial delay and what you might expect.
thanks for your reply.
I have already read your quotation. It helped me a lot. The speed of my disk is:
But the reason why I asked this question is that I want to know whether it is the real speed in my server,or is there something I’m missing. I am a new man in this field and will conduct in-depth research in my PhD career. So I don’t want to missing anything at the beginning.
I have found the fellowing data on your website:
But the speed in my server is as fellow:
my GPU is :4* Tesla P100-SXM2 16G .And my CPU is 128 * POWER8NVL.