hosted by Google, great
What could go wrong with my company's private data??
VMware is launching a cloud object storage service based on either the public Google cloud, or EMC ViPR for a private cloud alternative. It's called vCloud Air Object Storage, which amusingly acronymizes to vCAOS, not to VMware's liking we would think, and is intended for multi-petabytes of unstructured data. …
Well, that's why you should read the terms of service for vCloud Air and pay attention to the Data Privacy addendum, which applies to all data including that held on the Google service: https://www.vmware.com/support/vcloud-air/data-privacy-addendum.html
But I do understand if not bothering and posting FUD is easier.
only 99.9% availability for the best? From what I see that is almost 9 hours of downtime allowed per year. I'd figure more like six nines of availability at a minimum, with fancy multi site replication and multiple data centers, multiple ISPs etc and object storage that shouldn't be hard to achieve.
Cleversafe claims nine 9s of reliability (less than 32ms per year). Never used them but heard they are a leader in the space for on site object storage anyway.
Not sure how it compares to other object storage providers.
Do you have 6 nines of availability to the Internet? After all, you said it shouldn't be hard to achieve. That's about 3 seconds of total outages per year. That test will fail for an OSPF recalc, router restart, etc. Even a dropped NIC will typically result in more than 3 seconds of thinking about it before failing over.
If you really, really have 6 nines on your Internet connectivity, I'd like to know what you paid to put that in place. That's in front of paying peanuts for storage.
SLA From my isp(internap. Customer for 6 years) reads in part
"What our SLA covers*
Our SLA supports the following North American performance metrics:
100% network availability
Less than 45 milliseconds latency
Less than 0.3% packet loss
Less than 0.5 millisecond jitter"
Don't forget that the quoted "9"s availability should include planned and unplanned maintenance. With reasonable expectations of throughput and latency, you just cannot achieve 6 nines across the internet starting from CPE. Where do you draw the line of what is the client's responsibility and what is the providers? Does the client need to have access to multiple independent POPs, with redundant power supplies?
A telco might aim to achieve 5 nines end-to-end across their system (and report to the regulators when that fails). Do you really think an internet-based system is going go get close to that?
High nines is astoundingly expensive. Most don't need it.