So the Kinetic drives are internal hard disks with ethernet support built in? What is the use case for that? There isn't much of a market for a single-drive NAS, and the only thing I could see this drive enabling is cheaper single-drive NAS chassis.
Kinetic drives are meant for large scale object storage datacenters. The idea is that you can fill the server enclosure entirely with drives, with no need for a server motherboard/RAID cards to take up valuable space/money.
Indeed, large-scale object based storage is exactly what these are for, and 1Gbps switched Ethernet fabrics are far far cheaper than 6/12Gbps SATA/SAS expander/multipliers. Nothing to do with single-drive NAS.
Exactly, as Seagate likes to put it - isolation of storage from compute.
Traditional NAS / SAN tends to put the drives in 'compute servers' - so scaling the performance is a matter of putting in more powerful platforms - think of Atom-based NAS vs. Core-based NAS vs. Xeon-based NAS. With the Kinetic drives, the compute part and the drives / storage are separate - you can scale either depending on your needs.
That is really interesting. Can you still perform the same types of software raid solutions that you could with SATA drives? Being limited to 1 gbps would not be incredibly limiting for my home server, parity checks and disk rebuilds never exceed ~90 MB/s. File copies off server are already limited by 1 gbps.... so no problem there. Ethernet switches are hell of a lot cheaper than SATA cards.
It needs special software support from the compute side.. something like SwiftStack - basically deal with object-based storage instead of traditional file systems.
You'd use specialized software like Hadoop/HDFS or Ceph. Replication/Erasure Coding is then handled at the object level, as well as things like checksums etc. Obviously with disk sizes like this traditional block based RAID would yield pretty bad rebuild times otherwise.
I'm still curious what level of computing power they have on board. Even if they're currently only being targeted at the data-center; a single drive fileshare only NAS seems like an obvious future application for the tech. Unlike a media server/transcoder just serving up files wouldn't require much processing power.
Aside from the limitations of SMR, the Swift API uses http transport to transfer files, not your typical network filesystem protocol. The interface is also intended to read or write the entire file (or a very large chunk of a huge file). So, it's a good match for a media server backend, but not for random write applications. The Swift API is actually a good match for SMR technology. While I'm familiar with the Swift API, I'm not sure how the software is partitioned between the clients, the drives, and any required central permissions/index repository. Different partitioning leads to different implementation challenges.
SMR wouldn't be a show stopper (just put the controller on a different drive); not being able to talk standard file sharing protocols is seems like a bigger issue. At some point I suspect the basic hardware will get cheap enough someone will do it though, just like with the HDDs that have a USB controller baked directly instead of using sata to interface with the enclosure.
Slightly Off Topic, We have Toshiba and Intel stating 100TB SSD by 2020, Samsung's roadmap is slightly lower but there is no reason why they cant follow.
Seagate's Roadmap were 20TB by 2020. Price / GB for HDD will still likely be 3 - 4x cheaper then SSD ( Today we are HDD is 10x cheaper, assuming both SSD and HDD price dropping in the next few years )
But surely the additional benefits of SSD and TCO will come into play in some stage, and with 3D NAND and Xpoint, we may likely to see the shift in late 2016 or 2017.
How does Seagate intend to stay competitive. Will they speed up the Roadmap ?
Depends what you're looking at. In the data center, where space (volume), and power consumption (and fewer copies of data needed due to IO speed????) are a major chunk of the total cost flash is expected to overtake HDDs next year. OTOH a second graph near the very bottom of the study shows the raw price/TB staying in HDDs favor until 2023.
What we'll probably see is HDDs in the data center gradually get pushed from default frontline storage to archival data storage only. On the consumer side, I suspect the next year or two will see SSDs infiltrating the race to the bottom laptop segment; leaving HDDs almost exclusively secondary/media storage (either as a 2nd HDD in a desktop or in a NAS of some sort).
The bigger question, and one that the study I'm looking at seems to've ignored, is at what point will shrinking sales levels cut deeply enough into R&D that it slows the rate at which HDD capacity continues to grow. If it hits I could see that shifting the tipping point forward by several years. Once it hits, the HDD market will probably collapse very rapidly.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
14 Comments
Back to Article
Guspaz - Tuesday, September 1, 2015 - link
So the Kinetic drives are internal hard disks with ethernet support built in? What is the use case for that? There isn't much of a market for a single-drive NAS, and the only thing I could see this drive enabling is cheaper single-drive NAS chassis.takeshi7 - Tuesday, September 1, 2015 - link
Kinetic drives are meant for large scale object storage datacenters. The idea is that you can fill the server enclosure entirely with drives, with no need for a server motherboard/RAID cards to take up valuable space/money.qasdfdsaq - Tuesday, September 1, 2015 - link
Indeed, large-scale object based storage is exactly what these are for, and 1Gbps switched Ethernet fabrics are far far cheaper than 6/12Gbps SATA/SAS expander/multipliers. Nothing to do with single-drive NAS.ganeshts - Tuesday, September 1, 2015 - link
Exactly, as Seagate likes to put it - isolation of storage from compute.Traditional NAS / SAN tends to put the drives in 'compute servers' - so scaling the performance is a matter of putting in more powerful platforms - think of Atom-based NAS vs. Core-based NAS vs. Xeon-based NAS. With the Kinetic drives, the compute part and the drives / storage are separate - you can scale either depending on your needs.
3DoubleD - Tuesday, September 1, 2015 - link
That is really interesting. Can you still perform the same types of software raid solutions that you could with SATA drives? Being limited to 1 gbps would not be incredibly limiting for my home server, parity checks and disk rebuilds never exceed ~90 MB/s. File copies off server are already limited by 1 gbps.... so no problem there. Ethernet switches are hell of a lot cheaper than SATA cards.ganeshts - Tuesday, September 1, 2015 - link
It needs special software support from the compute side.. something like SwiftStack - basically deal with object-based storage instead of traditional file systems.iwod - Tuesday, September 1, 2015 - link
So this is like Software Defined Storage? Basically moving the Controller and Raid to Software and CPU?I wonder what the cable looks like? And if the HDD are powered through the same cable?
What File System is best suited for this?
Power Efficiency ?
nils_ - Monday, September 7, 2015 - link
You'd use specialized software like Hadoop/HDFS or Ceph. Replication/Erasure Coding is then handled at the object level, as well as things like checksums etc. Obviously with disk sizes like this traditional block based RAID would yield pretty bad rebuild times otherwise.DanNeely - Tuesday, September 1, 2015 - link
I'm still curious what level of computing power they have on board. Even if they're currently only being targeted at the data-center; a single drive fileshare only NAS seems like an obvious future application for the tech. Unlike a media server/transcoder just serving up files wouldn't require much processing power.jhh - Tuesday, September 1, 2015 - link
Aside from the limitations of SMR, the Swift API uses http transport to transfer files, not your typical network filesystem protocol. The interface is also intended to read or write the entire file (or a very large chunk of a huge file). So, it's a good match for a media server backend, but not for random write applications. The Swift API is actually a good match for SMR technology. While I'm familiar with the Swift API, I'm not sure how the software is partitioned between the clients, the drives, and any required central permissions/index repository. Different partitioning leads to different implementation challenges.DanNeely - Tuesday, September 1, 2015 - link
SMR wouldn't be a show stopper (just put the controller on a different drive); not being able to talk standard file sharing protocols is seems like a bigger issue. At some point I suspect the basic hardware will get cheap enough someone will do it though, just like with the HDDs that have a USB controller baked directly instead of using sata to interface with the enclosure.edward1987 - Friday, January 29, 2016 - link
128MB cache might indicate - working with large file- photo - video editing..you can ask these guys, they know: http://www.span.com/product/Seagate-Archive-HDD-v2...
iwod - Tuesday, September 1, 2015 - link
Slightly Off Topic, We have Toshiba and Intel stating 100TB SSD by 2020, Samsung's roadmap is slightly lower but there is no reason why they cant follow.Seagate's Roadmap were 20TB by 2020. Price / GB for HDD will still likely be 3 - 4x cheaper then SSD ( Today we are HDD is 10x cheaper, assuming both SSD and HDD price dropping in the next few years )
But surely the additional benefits of SSD and TCO will come into play in some stage, and with 3D NAND and Xpoint, we may likely to see the shift in late 2016 or 2017.
How does Seagate intend to stay competitive. Will they speed up the Roadmap ?
DanNeely - Wednesday, September 2, 2015 - link
Depends what you're looking at. In the data center, where space (volume), and power consumption (and fewer copies of data needed due to IO speed????) are a major chunk of the total cost flash is expected to overtake HDDs next year. OTOH a second graph near the very bottom of the study shows the raw price/TB staying in HDDs favor until 2023.What we'll probably see is HDDs in the data center gradually get pushed from default frontline storage to archival data storage only. On the consumer side, I suspect the next year or two will see SSDs infiltrating the race to the bottom laptop segment; leaving HDDs almost exclusively secondary/media storage (either as a 2nd HDD in a desktop or in a NAS of some sort).
The bigger question, and one that the study I'm looking at seems to've ignored, is at what point will shrinking sales levels cut deeply enough into R&D that it slows the rate at which HDD capacity continues to grow. If it hits I could see that shifting the tipping point forward by several years. Once it hits, the HDD market will probably collapse very rapidly.
http://wikibon.org/wiki/v/Evolution_of_All-Flash_A...