Clustered FS and Objectstore are built on top of SAS, FCOE, iSCSI, NVMe-OF. You first have to solve the problem of packing thousands of storage devices within the signal integrity radius of the transport medium before you can start abstracting. For NVMe that radius is about 1.5 - 2 meters from the CPU socket. SAS about 5 meters. Not sure on FC, I presume a couple km.

Completely common. Like 70% of all servers sold include a RAID controller that can talk to NVMe devices. But there's a catch... They suck so badly, nobody buys the PCIe cables to connect the backplanes. The inside joke is the best way to slow down your NVMe drives is to attach them to a RAID controller. Most NVMe drives use 4 PCIe lanes. Broadcom's RAID chips let them have two lanes. Then the RAID controller connects to the CPU with 16 lanes. So the minute you exceed 8 drives (via a switched backplane), you have an intractable bottleneck.

The Broadcom 3xxx chip hit the wall first as it still did RAID partially on the controller CPU. The 4116 implemented RAID entirely in silicon, the 51xx chips took this further with a complete cache redesign, and actually ditches SAS/SATA entirely. It's NVMe only. But nobody has solved the PCIe lane bottleneck.

Via a PCIe switch backplane. They've been around for a while... Perhaps as far back as 2012...

I fully expect SAS4 to be the end... NVMe-OF will replace SAS, and the drives will plug into the crazy 800GbE switches that are available now. Not on the drawing board... Now.

T

It's dead, Jim.

It's not coming back.

I got my hands on a 3B1 around 1990. Linux wasn't available of course, and a used Sun was just out of my reach. It was a giant box of suck. The most corporate version of Unix I've ever used. It made me pine away for the dialup guest account on Compupro's Unix Version 7 system in the east SF Bay back in the 80's...

Eventually I picked up a Sun 3/50, and later a 3/60. Then some kid in Finland posted on Usenet and well...

The whole point of Wordpress is so you can host your own site and not have to have your web site held hostage by one of the oligarch-controlled cloud providers!

No thanks. Good Bye!

Funding via pay per view? There's an Idiocracy joke in there somewhere...

I'm about done with NVIDIA. Here's to hoping Intel improves their Linux driver support... They step up their game, and I could be convinced to switch. But right now.... It's all crap.

T

100%

Kevin Rose killed digg. He took what was good and ruined it.

Reddit embodied what was good about Digg and ran away with the market share.

As long as Reddit doesn't fuck with their algo, and try to pull a "Kevin Rose" (not unlike what Tik Tok just did) it will be ok.

Fuck digg, and fuck Kevin. Go back to making videos where you pretend you're "edgy" because you have a 40oz in a brown paper bag.

Kevin and his idiot people killed Digg a long time ago when they implemented their own algo to determine what users could see and corrupted it with a bunch of advertising. It should have been left alone. Nobody cares and nobody wants it back.

So? How much single threaded rendering does anyone actually do?

I understand why in some cases single threaded performance is important, but not for the vast majority of use cases.

I don't actually know. My desktop runs Linux. I patch it a couple times a month perhaps, and grab a cup of coffee while it reboots.

But to the point of "Why hasn't it gotten faster", you have to understand what its doing. Servers have entire subsystems to inventory & boot, a modern Dell server has at least 3 OS'es (4 if you have an expander backplane) hiding under the hood before the one you see boots. So servers have a bit more stuff. But the fundamentals are the same for desktops & laptops. RAM and DDR training. Yes, PC's have gotten a lot faster. RAM has gotten a lot bigger, and more to the point we keep juicing more speed out of it. Getting DDR5 to hold on to the bus and emit a usable signal at 4800Mhz with signal integrity to travel 1 meter to the socket take a little bit of tuning.

Enjoy the cup of coffee.

T

The BSD's use something called Bhyve. I was pretty stable on TrueNAS Core when I used it, but not very feature rich. I only ran Linux VM's, so don't consider this an endorsement.

https://bhyve.org/

T

Clearing caches, ok... But... And maybe its because I'm just a BIND 9.x guy... How do I lower the TTL for a Microslop hosted record? Even for records I host, changing the TTL doesn't force the remote caches to expire early.

T

Which can basically undermine everything else at any point in the future.