Disk I/O speeds seem too slow
I am suspecting that something is wrong with the hard drives on my linode.
I am running a program that reads about 70k+ files when it starts. Usual time for startup for this program is around 30 seconds on a system with SSD. On this linode, it takes closer to 7 minutes. Writing files during shutdown for this program is also very slow.
Archiving files also seems extremely slow. I remember my linode being on a SSD. Am I wrong?
I remember this happening after I auto-upgraded the system from 20GB to 25GB. I didn't resize it in the system yet. I wonder if something went wrong there?
This may be due to having noisy neighbors that are using more of than their allotted resources. To check if this is the case run the following command when I/O is slow:
iostat 1 10
With this report we looking at the steal columns. This lets us know if your neighbors are using more of their shared resources than what allotted to them.
In a shared environment there's always going to be some steal. But as it increases you will start to see some latency.
If you are experiencing steal. Please open up a Support Ticket from your Linode Manager and send the output of the iostat command so we can take a look and see what is the best option.
Holy crap. I was on a node (since the free up-sizing) that was basically killing my server. I "resized" to a bigger server, then back down to the original size, which placed me on a different server, and now I'm no longer getting CPU use warnings and similar, despite alarms set at the same level. I didn't realize to check the "%steal" values before I moved, but h1101-fnc1 (freemont) was horrible. Now I'm on another node in freemont and "my" cpu use is practically nil and it's as responsive as I always remembered. Previously I was getting warnings daily for exceeding 200% CPU usage (all of it in I/O wait.)
To be clear, transfer out of that server was barely 20MB/s at peak, whereas transfer from 16GB back down to 8GB is going at around 120MB/s.
I love Linode, but they really need to figure out a way to balance I/O better. All the CPU share in the world means nothing if I can't open a file for several seconds because of contention with other VMs.
Seriously, the difference in performance between the original host (h1101-fnc1) and final host (h1229-fnc1) is night and day. And on that note, I really don't like how the new "cloud" UI doesn't show the host system a VM is on.