Some years back, I read this short PR style article[1] about how the LHC produces 10 petabyes of data per second. Which was (then) impossible to store in any meaningful way. So the folks at CERN came up with a clever triggering system that discarded 99.9+% of the data.
Now a wimpy 50 racks can provide minutes (well ~2.5 minutes) of buffer. A well planned 'university scale' data center could give them close to an hour. And a proper state-of-the-art hyper-scale data center would get dangerously close to a day. Wild.
It would be nice if you could eventually get huge SSDs as cheap as huge hard drives.
Like how you can get 30 TB hard drives for like $600.
rbanffy · 4h ago
These are not consumer grade devices, and I can't even find an easy quote for their 30TB devices.
The price point is determined by the benefit provided over the expected lifetime of the device, so, if you have a ~150 PB data footprint you'll needn't about 1 full rack of these drives (for 12 bays per rack unit, which is a low density).
To do the same with current HDDs, you'll get, at best, 360TB per rack unit, meaning you'll need roughly 10 full racks, a 10x increase in datacenter space. So, in space alone, you'll be spending 10x more with the HDD solution. If power and speed were the same (they aren't), this would mean that paying $48,000 for a drive like this would still be a good deal.
So, I assume it'll be a long time before we see a 245TB flash drive selling for $600.
metadat · 11h ago
Is the flash market a monopoly, or why is price per unit of storage so stagnant in recent years across both HDD and SSD/NVMe when this feat is possible?
Flash manufacturers not feeling an incentive to compete?
imtringued · 9h ago
Maybe you should actually look at the prices before complaining about the rapid decline cost of flash based memory.
Some years back, I read this short PR style article[1] about how the LHC produces 10 petabyes of data per second. Which was (then) impossible to store in any meaningful way. So the folks at CERN came up with a clever triggering system that discarded 99.9+% of the data.
Now a wimpy 50 racks can provide minutes (well ~2.5 minutes) of buffer. A well planned 'university scale' data center could give them close to an hour. And a proper state-of-the-art hyper-scale data center would get dangerously close to a day. Wild.
[1]: https://www.wired.com/2008/06/pb-quark/
Like how you can get 30 TB hard drives for like $600.
The price point is determined by the benefit provided over the expected lifetime of the device, so, if you have a ~150 PB data footprint you'll needn't about 1 full rack of these drives (for 12 bays per rack unit, which is a low density).
To do the same with current HDDs, you'll get, at best, 360TB per rack unit, meaning you'll need roughly 10 full racks, a 10x increase in datacenter space. So, in space alone, you'll be spending 10x more with the HDD solution. If power and speed were the same (they aren't), this would mean that paying $48,000 for a drive like this would still be a good deal.
So, I assume it'll be a long time before we see a 245TB flash drive selling for $600.
Flash manufacturers not feeling an incentive to compete?