Google Compute Engine Disks-First Glance

I’m currently working on getting Google Compute Engine up and running with Docker and a MySQL cluster. Coming from Amazon, there’s a few things that took getting use to.

1) Disk Space = IOPS

Unlike on Amazon where you can have a 100GB disk with 5000 IOPS, on GCE you get IOPS with size.  Want 5000IOPS? You’re going to need a big disk (5-10TB) and at least 4 CPUs.

More info here: https://developers.google.com/compute/docs/disks#pdperformance

2) Once you pick your disk size, you’re stuck

There is no easy way to go from a 100GB Disk to a 500GB Disk without just reinstalling and rebuilding everything. On Amazon you can snapshot and restore an EBS onto a larger drive, on Google your snapshot size is fixed.  So pick the LARGEST disk size you could possibly imagine needing. Luckily disk space is cheap (4 cents/GB), so it’s not terrible, but do plan ahead.

3) RAID is not needed

So far, I haven’t noticed a need to RAID my disks. GCE seems to do a good job of doing this for me, unlike the piss-poor performance you get from not RAIDing your EBS drives. Your load might be different, you might need a ton of IOPS and this is necessary, but for high-performance MySQL, it hasn’t seemed necessary from looking at my IOStats.

I highly recommend you check out Google Compute Engine, but its not for everyone. There’s still a lot to be desired and a lot that’s missing, but its worth exploring for your use case.

Leave a Reply

Your email address will not be published.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