Hacker News new | ask | show | jobs
by nullrouted 3879 days ago
Can you tell me why your disk latency and throughput is so horrible, even compared to AWS? I've seen ioping spikes in the 7-8 ms range and it has far worse throughput than AWS.

Also any idea if your folks are ever going to let people set RDNS records for their IPs? It has been requested but no one at GCE seems to communicate.
2 comments
larrymcp 3878 days ago
Oh wow... I didn't know Google lacks the ability for customers to specify reverse DNS. I had assumed any full-service cloud provider would offer this feature.

If Google doesn't have configurable RDNS, they're not a "one-stop shop" for all hosting needs. For example, a customer wouldn't send transactional e-mails from a server at Google because the lack of RDNS would affect deliverability.

link
boulos 3878 days ago
We are most definitely not a one-stop shop yet. We don't even have IPv6 on GCE (I recommend Linode for low traffic IPv6 to IPv4 bridges).

Moreover, (sadly) your example doesn't hold anyway: sending email is actually something all cloud providers try to avoid (SoftLayer in Brazil lately is the notable exception). So getting anyone to whitelist your smtp and other ports is usually an unpleasant experience.

link
larrymcp 3878 days ago
Oh, this actually works great at AWS, yeah! We've been sending transactional e-mails from our servers at Amazon for a couple of years. They do require you to request the e-mail ports be unblocked, but once we made the request it was approved within 24 hours. As a bonus, the same request form also has a space where you tell them what you want your reverse DNS to be. Very handy.

Last year Microsoft also announced Azure support for reverse DNS entries. So I would anticipate Google getting on this "bandwagon" fairly soon too. Just my guess though!

link
boulos 3878 days ago
Which "disk"?

We've got Persistent Disks aka PD (logically similar to EBS / network storage) that come in both a "hard drive" variety and an "ssd" variety, as well as local SSD.

I assume you're talking about one of the PD products, most likely PD-SSD and comparing to one of the EBS offerings (either the new one or a piops one)?

If so, unlike traditional EBS, PD scales throughput according to disk size [1]. You get 30 IOPS/GB up to 15k or so. As far as latency goes, what instance were you testing on? Your VM's I/O competes for cycles with your vCPU (this is true of all KVM-based virtualization), so if you spike up your CPU you can easily "starve" your I/O threads or vice versa.

As to your DNS question, I'll have to look into that. But where are you trying to communicate? (I mean clearly you found me, but I mean where else)

link
nullrouted 3878 days ago
Sure I use SSD Persistent disk and the latency is pretty horrible with the throughput being worse, it is much slower than a similar AWS EBS volume.

--- /dev/sda1 (device 20.0 GiB) ioping statistics --- 13 requests completed in 12.5 s, 161 iops, 647.9 KiB/s min/avg/max/mdev = 213 us / 6.2 ms / 18.2 ms / 6.0 ms

This is pretty bad, everything on AWS is under a millisecond.

I submit feedback through the feedback links in the panel and on the product forums. Since you folks don't offer any other way of communicating.

link