Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Rest in Peace, Optane

specbranch.com

Post reply on HN
Print view View original

91–100 of 122 posts

Re: Rest in Peace, Optane

#91

Post by RyanHamilton »

adastra22 wrote: post #86
zozbot234 wrote: post #67

Earlier quoted context omitted.

> mmap on Optane direct-access-aware (DAX) filesystems like EXT4/XFS now, is not like mmap on block devices where the OS gets in your way and pages stuff in from disk Yes, the real case for Optane memory is that, supposedly, you don't have to fsync(). And insisting on proper fsync() tends to tank the performance of even the fastest NVMe SSD's. So the argument for a real, transformative performance improvement is ther…

Why would you not have to fsync? The fsync is a memory barrier that is just as useful with octane to ensure integrity. Do you mean the latency of ensuring fsync safety is lower?

No you don't have to fsync. Think of it like RAM. You don't fsync RAM.

Re: Rest in Peace, Optane

#93

Post by xtracto »

hardware2win wrote: post #6

What was interesting about Optane was that it was kinda an attempt to get rid of disk and ram distinction. You had single device (optane stick) which served as both - ram and disk. After decades of current approach we finally could physically get rid of disks - hdds, ssds or even nvme sticks. https://www.theregister.com/2022/08/01/optane_intel_cancella... >Optane presented a radical, transformative technology but bec…

I was just thinking about optane now that I'm dealing with AWS lambda cold starts.

It wold be a great use for that technology to decrease function startup time .

Re: Rest in Peace, Optane

#94

Post by afdsganionio »

jandrewrogers wrote: post #29

Optane suffered the same fate as every other storage technology over the decades with a similar value proposition. Broadly, the notional value of storage technology like Optane is that it offers a drop-in improvement in storage performance without requiring any software modification, albeit at a significant cost premium. The theory that this market exists in any significant way never pans out in practice. On one hand…

>Broadly, the notional value of storage technology like Optane is that it offers a drop-in improvement in storage performance without requiring any software modification, albeit at a significant cost premium.

I don't think so at all. The value of Optane was that it was dramatically different, offering byte-addressable storage with latency of hundreds of nanoseconds. Getting the most out of this requires radically redesigning how modern systems store data, including ditching the idea of block storage.

Optane was the fastest traditional SSD on the market by a pretty wide margin, but that's not all it could have been.

>They use the storage so well that the gains by putting something in the middle will be marginal or non-existent, and certainly not worth the extra cost.

I don't agree with this either. How do you propose to get latency down to match Optane using ordinary SSDs? It doesn't matter how much hardware you throw at the problem; parallelism cannot reduce latency. You need another storage system to hit those targets. Maybe that can be a more traditional DRAM cache with battery backup, but that's only adequate in bursty workloads.

Putting "something in the middle" isn't worth it if the rest of the architecture stays the same. But it could bring about tremendous speedups if the new capabilities were used.

Re: Rest in Peace, Optane

#95

Post by ori_b »

xtracto wrote: post #93
hardware2win wrote: post #6

What was interesting about Optane was that it was kinda an attempt to get rid of disk and ram distinction. You had single device (optane stick) which served as both - ram and disk. After decades of current approach we finally could physically get rid of disks - hdds, ssds or even nvme sticks. https://www.theregister.com/2022/08/01/optane_intel_cancella... >Optane presented a radical, transformative technology but bec…

I was just thinking about optane now that I'm dealing with AWS lambda cold starts. It wold be a great use for that technology to decrease function startup time .

The problem is that if you want to keep the server state around, there's already a cheap and easy solution: write a persistent server.

Re: Rest in Peace, Optane

#96

Post by crabbone »

rcxdude wrote: post #64
hardware2win wrote: post #6

What was interesting about Optane was that it was kinda an attempt to get rid of disk and ram distinction. You had single device (optane stick) which served as both - ram and disk. After decades of current approach we finally could physically get rid of disks - hdds, ssds or even nvme sticks. https://www.theregister.com/2022/08/01/optane_intel_cancella... >Optane presented a radical, transformative technology but bec…

The problem is optane wasn't as fast as DRAM, nor as cheap as disk. So you still needed the conceptual split anyway, and there wasn't a compelling reason to really use it (OK, I can get a lot of RAM but it's 1/10th the speed of my normal RAM, which will kill both memory latency and memory bandwidth limited applications, which is basically any workload which wants lots of RAM. Or I can get super fast persistent storag…

Where Optane could've been used: distributed storage. The whole game there is about how fast can you ack the writes while writing as far away as you can. The company I worked for in this specific area had used pmem when that was available for that specific reason.

I'm no expert on costs of these things and especially wouldn't dare to predict how these costs could've behaved in the future, but my guess would be not that the technology had no use, rather that it was too expensive for what it was good at, and the second best was cheap enough to warrant choosing it over Optane.

Re: Rest in Peace, Optane

#97

Post by mdip »

choppaface wrote: post #8

Optane is dead. Long live Optane! PCIE 3.0 Optane recently sold so well it got backordered and the price popped. (Ok maybe it was LevelOneTechs Wendell? Maybe it was organic? https://www.newegg.com/intel-optane-905p-1-5tb/p/N82E1682016... ) Today PCIE 4.0 prices have fallen so fast that you can RAID0 a few M.2s and get about the same IOPs from a single Optane drive for a similar $ per GB, though not exactly the same…

It's possible the price pop is happening because of people trying to find spare parts.

I can't speak to these specific devices because I haven't been doing infrastructure work since before 2015, but I recall the price of certain enterprise products would routinely skyrocket a few years after it became hard to find them. I recall having to source a replacement drive for a server at my Dad's company and discovering that finding any drive at least 4.5GB in capacity for this 15 year old server was challenging. I ended up having to buy a much larger drive at about twice the cost of a modern (at the time) enterprise-grade SAS drive.

I doubt my specific example is all that similar to what might be happening. In my case, it was a confluence of the usual suspects: RAID controllers that are very particular about the specific drive models/firmware versions that they work reliably with (which everyone else needs to replace their drives, too!) coupled with the server using a no-longer-common interface (SCA) and a drive tray specific to the vendor which was no longer a popular (and went Bankrupt)[0].

I suspect it's too soon for it to be that simple but I wonder if there are specific circumstances where "My Optane drive failed and replacing this drive at any cost is the only fix that gets me access to my data, again".

[0] Not important, but it was a Gateway brand server if you can believe it. And I ended up kludging the tray situation, anyway.

Re: Rest in Peace, Optane

#98

Post by exmadscientist »

rwmj wrote: post #63

It's a real shame we can't force Intel to license it, and instead have to wait probably another decade for the patents to expire.

It's worse than that. As far as I am aware, the real problem with Optane is manufacturing cost, and lack of scaling up. So even if you have the patent rights, you probably still can't make it.

Re: Rest in Peace, Optane

#99

Post by adastra22 »

RyanHamilton wrote: post #91
adastra22 wrote: post #86

Earlier quoted context omitted.

Why would you not have to fsync? The fsync is a memory barrier that is just as useful with octane to ensure integrity. Do you mean the latency of ensuring fsync safety is lower?

No you don't have to fsync. Think of it like RAM. You don't fsync RAM.

You do, in fact. It’s called a memory write barrier. Ensures consistency of data structures as needed. And it call stall the cpu pipeline, so there’s a nontrivial cost involved.

Re: Rest in Peace, Optane

#100

Post by lstodd »

jiggawatts wrote: post #41
lstodd wrote: post #38

Earlier quoted context omitted.

Yeah, this. Intel was all talk and zero samples, not even a demo. So back then we just shrugged and moved on.

Literally unoptaneium?

Pretty much.

I remember sitting through those pitches and being like "Ok, you don't have any idea how and where to apply this new thing of yours, which is understandable, but then you are unwilling to let anybody have a shot at it. What I'm doing here then?".

IDK what they were thinking, but the general mood was "oh, another Itanic".

Post reply on HN