(Disclaimer, I work for Google, but these comments are my own observations.)
For what it's worth, there are workloads where e2 is a win --- where using an e2 VM is faster than n1 and cheaper to boot. Using gce-xfstests[1], which is a kernel regression test suite runner for file systems. It's mostly disk I/O bound, but it does have enough CPU that e2 was a win.
(Cost includes a 10GB PD-Standard OS disk and 100GB PD-SSD test disk.)
Note that for this particular use case, the sustained use discount isn't applicable, because it's not something which is continuously running. I'll make changes to the kernel and either run the auto group, which takes a bit over an hour and a half, or run the quick group, which will give me results in around 20 minutes.
5 cents savings per run might not seem like a big deal, but I tend to run a lot of tests, since a command like "gce-xfstests smoke" plucks the built kernel from my sources, launches a test appliance VM, and then results will land in my inbox a short while later.
For more complete testing, (especially before I send a pull request upstream), I'll use the Lightweight Gce-xfstests Test Manager (or LGTM) to automatically launch 11 VM's to run the auto group in a large number of different fs configurations. So a typical test VM is only going to be running for 2 hours (plus or minus), and so it's very bursty. But isn't that one of the best ways to use the public cloud?
Example e-mail'ed report from gce-xfstests using the LGTM runner:
(Disclaimer, I work for Google, but these comments are my own observations.)
For what it's worth, there are workloads where e2 is a win --- where using an e2 VM is faster than n1 and cheaper to boot. Using gce-xfstests[1], which is a kernel regression test suite runner for file systems. It's mostly disk I/O bound, but it does have enough CPU that e2 was a win.
[1] thunk.org/gce-xfstests
Command-line: gce-xfstests -c ext4/4k -g auto --machtype
(Cost includes a 10GB PD-Standard OS disk and 100GB PD-SSD test disk.)
Note that for this particular use case, the sustained use discount isn't applicable, because it's not something which is continuously running. I'll make changes to the kernel and either run the auto group, which takes a bit over an hour and a half, or run the quick group, which will give me results in around 20 minutes.
5 cents savings per run might not seem like a big deal, but I tend to run a lot of tests, since a command like "gce-xfstests smoke" plucks the built kernel from my sources, launches a test appliance VM, and then results will land in my inbox a short while later.
For more complete testing, (especially before I send a pull request upstream), I'll use the Lightweight Gce-xfstests Test Manager (or LGTM) to automatically launch 11 VM's to run the auto group in a large number of different fs configurations. So a typical test VM is only going to be running for 2 hours (plus or minus), and so it's very bursty. But isn't that one of the best ways to use the public cloud?
Example e-mail'ed report from gce-xfstests using the LGTM runner: