As Google enters the era of clouds "Recommend evolution is necessary for HDD" Recommendation

ByKamil Porembiński

The common sense that has been thought to be "the HDD is easy to break down if the temperature is high or the access frequency is high" is calculated from data using over 100,000 consumer HDDsHDD fails regardless of temperature or frequency of useGoogle, which has been repainted, is now researching HDDs for data centers and makes recommendations based on the results.

Google Cloud Platform Blog: Google seeks new disks for data centers
http://googlecloudplatform.blogspot.jp/2016/02/Google-seeks-new-disks-for-data-centers.html


Disks for Data Centers
https://research.google.com/pubs/pub44830.html

As of 2016, many of the HDDs on the market are optimized for standalone operation and are not assumed to be treated as "a part of disk aggregate" at a place like a data center.

There are a number of things that Google needs to have a large data center. For example, YouTube is one of which a lot of movies are posted. Below is a graph "How much data is uploaded to YouTube?", The horizontal axis shows time lapse, and the vertical axis shows how many hours of data are uploaded per minute .


Although it can be read as if it is growing steadily to the right, it is ready to grow steadily, but in the horizontal direction is 2 years increments, while the vertical is "1" "10" "100"Semilogarithmic graphPlease note that it is. In other words, the amount of uploaded data increases exponentially with each year. As of 2016, the data uploaded in 1 minute is equivalent to 400 hours, 1 GB is required for 1 hour of this data, so in the data center every day 1PB (petabyte)It is a state where expansion is necessary. this is,Library of Congress100 times the amount.

Since it is digital data, there is no problem if copies remain, but it is better if data loss does not occur. Nonetheless, revenues will increase if you are expanding the capacity and effort to avoid data loss, toward capacity expansion and system performance improvement. So Google insists that it is necessary for vendors, researchers, and consumers who need HDD to join together to create the "data center disk" necessary for the cloud era .

What we are looking for is
·IOPS(Number of read / write times per second) is high
· Large capacity
·tailLatencyLow value
· Meet security requirements
·Total cost of ownership (TCO)Is low
So, there are not many demands, including companies that operate huge data centers other than Google. As the optical disc has evolved into CD, DVD, Blu-ray Disc, it may be now that the HDD evolves fundamentally.

· Continued
What is an important factor in estimating the reliability of SSDs found by studying SSDs used in Google's data center? - GIGAZINE