ARMv8 in the Datacenter – 96 physical cores 0.50/hr | Hacker News Reader

ARMv8 in the Datacenter – 96 physical cores 0.50/hr

packet.net

39 pointsthepumpkin197913 comments

For those of you asking about Packet reliability - been a huge fan of them and supported them where I can. My blog post and company have been on the front page of HN before. Here are my benchmarks: http://blog.tiingo.com/switched-away-aws-packet-net-benchmar...

Since I've been with them, I haven't experienced any downtime except for block storage where I knew getting into it that it was in alpha at that time and I was an early tester. Even that downtime was at 99.9% and they had engineers on the phone with me throughout the night updating me even though it was only down for a couple hours. They don't charge me extra for support which was nice.

I ended up deciding to use their NVME storage for faster performance.

I can't recommend them enough. The backend of Tiingo would not be possible without them and their co-founders and engineering staff helped me quite a bit when I was looking for an AWS replacement. Took a chance with them a year ago (before Series A) and don't regret it.

I went to them initially as I have market feeds coming from NY4/NY5 and AWS network throughput wasn't good enough. Data was delayed and inconsistent. Also Packet had a data center in NJ, 25ish miles from the market data center whereas AWS had it in Northern VA which was hundreds of miles.

Right now I have real-time market streaming and the only downtime I've experienced has been user error. I usually have the #packethost freenode channel up if anybody wants to ask me directly (rishiattiingo)

Hope this is helpful

While the pricing is compelling, most customers want at least 98℅ of the reliability that industry giants offer. Maybe the community can comment on whether packet.Net is reliable enough to trust your business with?

FYI I'm from Packet. Agreed reliability is important (I'll let others comment on our own record) but I'd say there is significant workload via spot market, batch processing, etc that doesn't need uptime and is interested in cost.

Also, I've seen most users want 99.9+% uptime. 2% downtime is 14 hours in a short month. People would probably get angry about that...

Hey

Any chance you guys are going to offer something that is on the lower end of the spectrum for ARMv8 in terms of pricing? For eg. Scaleway has ARMv7 servers for 3 EUR/month (https://www.scaleway.com/pricing/)

Yes, we are considering a Type 0 ARM server, but we are still working through the proper design to enable full Armv8 (vs v7).

Sweet :D

I'll keep an eye on the Packet website. I did try out the ARMv8 server, a bit disappointed that the hardware did not support backwards compatibility with ARMv7 though :(

Looking for something similarly priced to scaleway in US too. Have a ton of services with them, but can't find an alternative at lower latency.

At CoreOS, we use Packet to host the Quay.io build system[0]. Getting bare-metal machines from Packet that we can run our own virtualization on has saved us a significant portion of our expenses compared to our previous system which dynamical spun up EC2 instances. This is definitely not the traditional deployment, but if you're using things like dedicated servers on AWS, you should be evaluating Packet.

In my own personal ops experience with Packet, I've actually seen less failures than Amazon; it feels like more than once a week Amazon has degraded service on the EC2 API in us-east-1 that used to cause us downtime. Not to mention, Amazon waits like 4 hours before they admit anything ever went wrong (green with a tiny yellow badge). However, I'd take this with a grain of salt, as we've replaced the Amazon control plane with a Kubernetes control plane and only interact with Packet API when we tear-down/spin-up a new cluster.

[0]: https://blog.quay.io/faster-container-builds

we use packet.net for all of our services at andyet.com and i can only remember one time that we had a problem that was actually because of them and not user error. as i recall they had an issue with some networking hardware, but had it repaired in about an hour.

i've seen a few other times where they've mentioned issues in their status page, but they're typically only for low inventory, and even those are usually resolved pretty quickly.

overall we're super happy with packet and i recommend them to anyone looking for hosting - especially those that have a need for bare metal servers.

I checked out the type 0. It seems to be about twice as much memory as the digitalocean one for roughly the same price. 8GB for 40$ on Type 0. But, I would really need to investigate it to make sure that we can upgrade and get a similarly good deal. Also, the ease of upgrade, as well as uptime would need to be investigated too.

Sort of asked before, but does anybody have experience with packet.net as a cloud host?

Already out of stock! That's annoying, I was all set to try it out.

we managed to get more online -- please go ahead and try again!

Reply on news.ycombinator.com