3 ms·

Few people have a problem that Optane directly solves. Its utility is narrow in the best case and you would need to specifically engineer a software stack to ta

Few people have a problem that Optane directly solves. Its utility is narrow in the best case and you would need to specifically engineer a software stack to take advantage of it for some small part of the memory address space. In many cases it is desirable for memory to be volatile. If high-throughput paging mechanics are required, like databases, this technology doesn't serve much of a purpose. Storage densities per server have only been increasing. It has some obvious utility for transaction logs and such but falls into a notorious dead spot in the market -- people with low volumes don't need the performance and people with high volumes can achieve similar results by other means for a price that isn't that different from the Optane premium. I've tried to find a use for it, and I haven't been able to. That doesn't mean a use case doesn't exist, but if it was "the biggest new idea" you'd think the use case would be more obvious for more applications.

zogomoox 4y ago
The use case is replacing flash, because modern flash cells die after a handful of writes. Surely there's a market for ultra-durable SSDs.
- jandrewrogers 4y ago
  That math rarely pencils out. Most companies with large storage budgets have figured out that you can design write-intensive storage engines for cheap read-optimized flash that will last several years without an issue despite the relatively low number of write cycles. This is easy to verify on a cocktail napkin if one can assume the software is efficient with its use of storage bandwidth. In most applications with sane storage architectures, you might physically rewrite an SSD once a month, and often less, despite relatively high write intensity. Quite a lot of software is excessively wasteful with storage writes, in which case it is more of an issue. That says more about the software than the storage, and can be addressed by not being profligate.
  - anonymousDan 4y ago
    What do you mean by 'wasteful' with storage writes? That there is some unintentional write amplification going on for whatever reason?
tester756 4y ago
>you would need to specifically engineer a software stack to take advantage of it for some small part of the memory address space. didnt they already do that in kernel and/or drivers? because Optane had customers, that wasn't some crazy lab tech, it was actually deployed in prod envs.

zogomoox 4y ago

The use case is replacing flash, because modern flash cells die after a handful of writes. Surely there's a market for ultra-durable SSDs.

jandrewrogers 4y ago
That math rarely pencils out. Most companies with large storage budgets have figured out that you can design write-intensive storage engines for cheap read-optimized flash that will last several years without an issue despite the relatively low number of write cycles. This is easy to verify on a cocktail napkin if one can assume the software is efficient with its use of storage bandwidth. In most applications with sane storage architectures, you might physically rewrite an SSD once a month, and often less, despite relatively high write intensity. Quite a lot of software is excessively wasteful with storage writes, in which case it is more of an issue. That says more about the software than the storage, and can be addressed by not being profligate.
- anonymousDan 4y ago
  What do you mean by 'wasteful' with storage writes? That there is some unintentional write amplification going on for whatever reason?

tester756 4y ago

>you would need to specifically engineer a software stack to take advantage of it for some small part of the memory address space. didnt they already do that in kernel and/or drivers? because Optane had customers, that wasn't some crazy lab tech, it was actually deployed in prod envs.