![]() The article text states: "These devices form the basis of in-memory computing: an approach in which both information processing and storing computational data are performed on the 10 same physical devices organized in a computational memory unit. We experimentally demonstrate the efficacy of the approach by accurately solving systems of linear equations, in particular, a system of 5,000 equations using 998,752 phase-change memory devices. The system therefore benefits from both the high precision of digital computing and the energy/areal efficiency of in-memory computing. In this hybrid system, the computational memory unit performs the bulk of a computational task, while the von Neumann machine implements a backward method to iteratively improve the accuracy of the solution. We introduce the concept of mixed-precision in-memory computing, which combines a von Neumann machine with a computational memory unit. Nature Electronics, 17 April, ran an IBM Research paper, Mixed-Precision In-Memory Computing, which talked about using Phase-Change Memory (PCM) in a compute-in-memory application. If not, it's going to be hard to argue for it. If Pelican costs less in capex, and less in operating costs, the archiving use case could be a reasonable one. So, if capex goes up by 50 per cent, you'd have to get the data centre space and power for free to break even. Operating Costs: Data centre space and power account for about a third the cost of the capex in a cloud storage system.(About halfway through this post (Storage Pod 6.0), you can see the $/GB in capex for each model of hardware we had at the time obviously lower now.) CapEx: Ask Microsoft what the capex for Pelican is.However, without numbers from Microsoft, it's hard to say whether that is true. There are use cases (primarily archiving) for such a technology if it really is very inexpensive. Replicating a model where it's difficult to get the data with hard drives isn't necessarily a solution. We've looked multiple times over the years at storing data on tape, but in a cloud environment where customers want to be able to access the data, it never penciled out. ![]() However, storing data that you won't be able to use – either because it's too slow or too expensive to get at – is much less useful. Technologies that help drive down costs help us all. Gleb Budman As you know, Backblaze believes strongly in driving down the cost of storage, and that lower-cost storage enables new use cases and businesses. However, object storage is generally not as fast as block storage.Įl Reg Is there a real need (that Backblaze sees) for a storage technology that is faster than tape and slower than disk, but nearer tape cost than disk cost? If there is then would Pelican meet that need? Object storage also supports having metadata for objects. In general object storage is designed for scale in a way that file and block storage systems are not (we currently store billions of objects and are on-track to have an exabyte of storage this year). Our cloud backup service uses that underlying our B2 cloud service exposes that via APIs. Gleb Budman Yes, the Backblaze cloud storage is an object store. Does Backblaze use object storage? What are the arguments for and against the storage technology you use? write once, rewrite never, read almost never) that's probably a reasonable choice.Įl Reg Pelican is using object storage (talks of storing blobs and using Gets and Puts). The SMR disks are priced lower, so that's compelling, and if the use cases are literally replacing tape (i.e. The Linux kernel is starting to add workarounds for some of this with smart filesystem bits, but they're still very early on. I'm sure Microsoft has found that out in their evaluations. Gleb Budman SMRs have really bad rewrite performance, and would require some pretty major software/hardware tweaks to use effectively. What are your views on using shingled and non-shingled disks? I doubt most cloud-scale companies would, but enterprise IT folks may.Įl Reg Pelican is evaluating shingled and non-shingled disks. Having said that, some users may have use cases to fit this pattern. However, between users wanting to access certain files and the system doing housekeeping, it hasn't made sense.įor our cloud storage service, B2, we also wouldn't want to limit the fundamental access to data since it's a real-time service.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |