HDD Performance – Part 6 – 16 TB Write then Read

Write performance in MB/s – 16 TB Seagate Ironwolf, avg. 198 MB/s

Wow, this is part 6 of the series already, but I keep getting interesting results. One of the questions that came up during part 5 was how fast data could be read from one of the newer and faster drives when the drive is almost full. To answer this question I ran the following test: Write 16 TB of 50 GB files until the drive is full, then immediately read the drive again. Reading files with a size of 50 GB each should show the maximum average read speed over the drive because there should be lots of chunks for 50 GB on the drive that are consecutive. So here are the results:

The first image on top is straight from part 2 of this series and shows the write performance of a 16 GB Seagate Ironwulf drive. The average write speed of the drive was 198 MB/s. For details have a look at that part. And here is the graph resulting from reading the 16 TB of data:

Reading 16 TB of data in chunks of 50 GB files

Now this graph looks so much different from all other graphs in this blog series so far. I would have expected that reading the files would be done in pretty much the same order as they were written, but this is obviously not the case. The speed values over time are all over the place with an average read speed of 206 MB/s across the complete HDD.

From my point of view this is a pretty good value. However, in my typical use case, I have super large, big and small files, so overall performance when the drive is full will probably not reach 206 MB/s. My estimation is that it would go down to perhaps 150-170 MB/s, which is not much faster than the average read speed of my almost full 8 TB drives, which I tested in part 5 of this series, which came out with an average speed of 121 and 145 MB/s.

Next…

So much for today. In the next installment, I’ll have a look at the performance of my different backup hard disks when I do a differential backup. These are quite large, because the huge virtual machine files usually have to be fully replaced, and there are also lots of changes with the large and small files. This is an important data point for me, because differential backups should complete as quickly as possible.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.