Anybody Using AV Linux?

[-Asdfgh-]

It's a BIOS level RAID. AFAIK it would be driver independant. I should probably update the firmware before giving up on it.

Nothing I do is important enough to back up

Linux may still need the correct drivers. I wrongly assumed that because the RAID on my desktop was enabled in the BIOS which then enabled the ability to group the disks in a little BIOS utility that it was equivalent to a hardware RAID but it was not. I might go and fix it at some point as it will give me more overall resilience to a single disk failure, but the real pain becomes backing up Windows in its entirety and reinstalling it after setting up Linux. Saying that, with OEM windows backing up at some level is essential.

 

[BadExample]

I'm glad you explained that. I'll skip BIOS update and just take them out of the RAID altogether. Not much point in a JBOD RAID in Linux anyway since it is normally multi-partition.

but the real pain becomes backing up Windows in its entirety and reinstalling it after setting up Linux. Saying that, with OEM windows backing up at some level is essential.

I use ghost, image the entire disk, if you have space for it on an external drive. I haven't used this one since the 90's, but I hear it is very refined and just as good as ghost: Clonezilla - About What sucks about any kind of backup is how do you test it if you don't happen to have identical hardware laying around unused. Not common in even enterprise level IT. I've been burned. Actually my users have been burned If you have a spare hard disk, it would be good to test a recover to that disk with the original pulled out or unplugged.

I don't know when I'll get time to play again

 

[-Asdfgh-]

I'm glad you explained that. I'll skip BIOS update and just take them out of the RAID altogether. Not much point in a JBOD RAID in Linux anyway since it is normally multi-partition.

RAID and LVM are great together, although to be fair I have not tried LVM in a dual boot situation.

I haven't used this one since the 90's, but I hear it is very refined and just as good as ghost: Clonezilla - About

I use this at home, and have at work sometimes, although work is based more on provisioning hundreds if machines, so clonezilla isn't really the ideal tool for that. You can clone to various types of media, though.

 

[BadExample]

I've setup the ghostcast server/suite a couple of times but always end up going back to imaging one at a time. It seems there is always enough post image fiddling around to not be very helpful to me to have a stack of pre imaged PCs to roll out. I have less PCs out there too. Maybe 120ish. I'm in a really weird paradyme. It usually doesn't make life any easier for me if I speed up the process. New computers are more or less the assgnats of my job.

 

[-Asdfgh-]

I've setup the ghostcast server/suite a couple of times but always end up going back to imaging one at a time. It seems there is always enough post image fiddling around to not be very helpful to me to have a stack of pre imaged PCs to roll out. I have less PCs out there too. Maybe 120ish. I'm in a really weird paradyme. It usually doesn't make life any easier for me if I speed up the process. New computers are more or less the assgnats of my job.

Check out something like Puppet (and Razor) for work.

 

[somebrains]

Has anyone tried building out their dependencies in Docker?
Orchestrating your audio workflow with Kubernetes can't be nearly as resource dependant as some of the ETL streams we run at work.
Basically, I wonder if you can break streams of audio work, swarming as you need, where you need it if you had to run an optimized environment.

Since this is for yourself, you wouldn't have to worry about container security as far as sharing volume mounts or granular user permissions.

Just built your images, test, chain out the best performing pod.
When your chain has to run into a bare metal machine export the file, take it back up as tested and appropriate.

My workstation is an E3-1270 with 32gb ram.
I'm running a single ssd for win10 and a 3tb drive for hyper-v.
I can crank a serious Docker swarm, into Rhel7 VM's or bare metal Windows (rarely needed).
We push our builds out to either AWS, or our own racks of HP Gen7 large build hosts.

Power is never the issue, bc I can buy a bunch of it cheap.

The issue is a workflow that requires a chain of builds that aren't the old single image baked into Chef that are big iron clones.

I need lots of little containers processing and pushing on data, eventually to a big box (sometimes).

Food for thought.

 

[-Asdfgh-]

Has anyone tried building out their dependencies in Docker?
Orchestrating your audio workflow with Kubernetes can't be nearly as resource dependant as some of the ETL streams we run at work.
Basically, I wonder if you can break streams of audio work, swarming as you need, where you need it if you had to run an optimized environment.

Since this is for yourself, you wouldn't have to worry about container security as far as sharing volume mounts or granular user permissions.

Just built your images, test, chain out the best performing pod.
When your chain has to run into a bare metal machine export the file, take it back up as tested and appropriate.

My workstation is an E3-1270 with 32gb ram.
I'm running a single ssd for win10 and a 3tb drive for hyper-v.
I can crank a serious Docker swarm, into Rhel7 VM's or bare metal Windows (rarely needed).
We push our builds out to either AWS, or our own racks of HP Gen7 large build hosts.

Power is never the issue, bc I can buy a bunch of it cheap.

The issue is a workflow that requires a chain of builds that aren't the old single image baked into Chef that are big iron clones.

I need lots of little containers processing and pushing on data, eventually to a big box (sometimes).

Food for thought.

For infiniband the network latency overhead is negligible for docker, and I presume the same would hold for audio, but Linux audio systems aren't exactly the best, so I'm not sure how well it would work in terms of contention for hardware resources. If it's just passing through data streams it should work, but given that I run Reaper with VSTs, and even given Reaper's ability to offload processing, I'm not sure I could build a convincing use case for the complexity.

If you are in the UK and I'd seen your post week ago I might have offered you a job, though!

 

[somebrains]

I used to live in Hollywood, and drank coffee with the Native Instruments devs every morning. A bunch of us used to throw ideas against the wall bc the nature of data is under-critiqued.

Today, we have stoppage from lack of vision of any architecture being inherently the same as a traditional datacenters. Many builds do not leverage networking, storage, and bursty/chuggy compute as the build target.

Data seems to be immutable, but I build out a VMware 5.5 cluster for a B2E ISP. I really saw my network admin think, and work, and tier. He is more what we think of as Sys engineer or DevOps architect. He got behind the Cisco and Juniper hardware and we went to down a virtualization rabbit hole earlier than a lot of carrier class businesses. EMC told us flash SANs weren't viable, so we built them as edge accelerators. S come rev we built replacements as iSCSI core storage. About a year later AWS made their switch and EMC pushes flash iSCSI SANs. Juniper told us that virtualized 40Gbit switching wasn't market viable, then they bought ubnt after we built our stack out of Dell hosts and out it in production. They're using their Asics on basically a commodity VM host and a massive orchestrated swarm of router instances now. Totally normal, everyone should be doing it.

You can treat data like soda thru a crazy straw, but a lot of people see data as immutable to a degree, and deployments as "immutable architecture". That's the problem with AWS useage. If you get too reliant on their managed services no one bothers to build assumption until there's a major outage. Then you are too late bc everything in flight rarely is loosely coupled to Redshift or Kinesis. The old way was running M class instances across multiple accounts multi region multi zone with DNS switches waiting to fire from micro instances running as listeners. Now you just hope your forced maps work bc you don't have that control using majority services.

So we build and test, and rebuild, and test, and test, and test.

Always looking to address the dark scary things we aren't supposed to question, or mess with.

Devs specifically don't like us choking, splitting, or accelerating data for infrastructure reasons.