Oxide
-
@warren-stanley said in Oxide:
@scottalanmiller there's a bit of a rabbit hole to tumble down on this. The topics Jessie posts tend to be Security and Open-Source oriented with interesting, but not limited to, views on the BMC and IPMI tech we regularly use. I've started reading through the back catalog posts of the other people involved to see if i can get a handle on what they're aiming at.
Yeah, the problem there is like me... I "tend to talk" about certain things. But then my commercial projects are often unrelated. People are constantly confused by what I'm working on because they tend to see one thing that I talk about and think that that is what (and probably all) that I do. Then they think that that implies that everything else that I do is related to that thing.
This might be nothing to do with security, open source, IPMI, etc. Or it might totally be. The site says... nothing at all. I'm not even sure of the fundamentals like... do they plan to manufacture something?
-
Well, it says what it is, kind of.
Rack-scale design (RSD) is a way to build dynamic, software designed datacenters. Intel is behind it with partners like HPE, Dell, Supermicro etc building hardware for it. It's an architecture aimed at hyperscale datacenters with the purpose of higher resource utilization and lower cost.
It's not just hardware but software as well. Basically you want to disaggregate everything - cpu, memory , storage, network, accelerators. And being able to upgrade and scale this independently. The buzz word for it is Composable disaggregated infrastructure (CDI). It looks like it's a blade architecture when you get into details but it looks like the smallest building block would be a rack.
They (Oxide) says they are bringing this technology to on-prem for everyone. So I assume they are making it more accessible and more turnkey. Considering how many people and resources are required for hardware, I say they are doing software only.
Look for Intel RSD to find more info.
Dell's DSS-9000
https://www.storagereview.com/dell_emc_dss_9000_rack_scale_infrastructure_announced
-
@Pete-S I'm really hoping that the on-prem for "everybody" is implying scaling down to a very small size
-
@warren-stanley said in Oxide:
@Pete-S I'm really hoping that the on-prem for "everybody" is implying scaling down to a very small size
If you scale it down, does it stop being Rack Scale, though?
-
@Pete-S said in Oxide:
Well, it says what it is, kind of.
Rack-scale design (RSD) is a way to build dynamic, software designed datacenters. Intel is behind it with partners like HPE, Dell, Supermicro etc building hardware for it. It's an architecture aimed at hyperscale datacenters with the purpose of higher resource utilization and lower cost.
It's not just hardware but software as well. Basically you want to disaggregate everything - cpu, memory , storage, network, accelerators. And being able to upgrade and scale this independently. The buzz word for it is Composable disaggregated infrastructure (CDI). It looks like it's a blade architecture when you get into details but it looks like the smallest building block would be a rack.
They (Oxide) says they are bringing this technology to on-prem for everyone. So I assume they are making it more accessible and more turnkey. Considering how many people and resources are required for hardware, I say they are doing software only.
Look for Intel RSD to find more info.
Dell's DSS-9000
https://www.storagereview.com/dell_emc_dss_9000_rack_scale_infrastructure_announced
Sounds a lot like the SGI servers from the late 90s.
-
@scottalanmiller said in Oxide:
@warren-stanley said in Oxide:
@Pete-S I'm really hoping that the on-prem for "everybody" is implying scaling down to a very small size
If you scale it down, does it stop being Rack Scale, though?
I can imagine a pretty small rack
-
@warren-stanley said in Oxide:
@Pete-S I'm really hoping that the on-prem for "everybody" is implying scaling down to a very small size
A hyperscale datacenter is by definition at the very minimum 5000 servers. So a single rack is indeed everything scaled down to a very, very, tiny small size.
If you can run your entire workload on just a handful of hosts...you're probably not "everybody".
-
@Pete-S said in Oxide:
If you can run your entire workload on just a handful of hosts...you're probably not "everybody".
The average company is just one or two servers. Really, if you need more than a handful of hosts, you are already in a decently exclusive category.
The problem being "rack scale" and "everybody" are exclusive concepts. I expect that this is trying to bring rack scale ideas to sub-rack scale size, but even so, composable infrastructure is for rather large shops as a starting point.
-
@scottalanmiller said in Oxide:
@Pete-S said in Oxide:
If you can run your entire workload on just a handful of hosts...you're probably not "everybody".
The average company is just one or two servers. Really, if you need more than a handful of hosts, you are already in a decently exclusive category.
The problem being "rack scale" and "everybody" are exclusive concepts. I expect that this is trying to bring rack scale ideas to sub-rack scale size, but even so, composable infrastructure is for rather large shops as a starting point.
LOL - maybe they mean everybody that's a F100..
-
She wrote a post on her blog about it that explains it a little better.
-
From her past posts, I'm assuming it's going to be something aligning with open compute and taking things like how Facebook designed their racks/servers into account.
Like here:
https://twitter.com/jessfraz/status/1106336080956018689?lang=en
