Mechanical hard drives have saturated the transfer rate of SATA3 since they moved to 256MB buffers years ago, they just don't do it for more than a second or two. This drive will actually be close to SATA3 limits during short burst transfers - the same way SSD's have been before going to NVMe, due to its large buffer AND incredibly high sustained transfer rate. This is yet another reason you see SAS used on enterprise drives, especially when a RAID is used.
But I get the joke you're implying, it's just important to keep in perspective the majority of workloads for these drives is burst, not sustained transfers ;)
Yes, that cache works for shorts bursts. But the chance of hitting the cache for reads is very low, especially if you consider that the OS already uses GBs of "free" DRAM as read cache.
I still wonder why we don't have mixed drives. 10 TB HDD for everything 1MB or bigger. 512 GB SSD for everything 1MB or smaller and indexing HDD. Small does not clog IOPS on HDD, big does not fill SSD. search time is at SSD speed, capacity is enough for most usecases.
Something similar is called ZFS and caches on a SSD (L2ARC) the files you actually use, so that you can get 500 MB/s (or whatever your SSD does) even on big ones, without wasting SSD for the many but rarely used small ones.
Not exactly what you asked, but AFAIK the best compromise so far.
I wonder. How about moving whole MBR to MMR. It's nonvolatile, fast and you could save a lot on latency when heads don't have to seek where the data is stored.
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
8 Comments
Back to Article
dgingeri - Wednesday, July 24, 2019 - link
So, mechanical hard drives have finally gone beyond the transfer rate of 3Gb/s SATA.Samus - Wednesday, July 24, 2019 - link
Mechanical hard drives have saturated the transfer rate of SATA3 since they moved to 256MB buffers years ago, they just don't do it for more than a second or two. This drive will actually be close to SATA3 limits during short burst transfers - the same way SSD's have been before going to NVMe, due to its large buffer AND incredibly high sustained transfer rate. This is yet another reason you see SAS used on enterprise drives, especially when a RAID is used.But I get the joke you're implying, it's just important to keep in perspective the majority of workloads for these drives is burst, not sustained transfers ;)
MrSpadge - Wednesday, July 24, 2019 - link
Yes, that cache works for shorts bursts. But the chance of hitting the cache for reads is very low, especially if you consider that the OS already uses GBs of "free" DRAM as read cache.deil - Wednesday, July 24, 2019 - link
I still wonder why we don't have mixed drives.10 TB HDD for everything 1MB or bigger.
512 GB SSD for everything 1MB or smaller and indexing HDD.
Small does not clog IOPS on HDD, big does not fill SSD.
search time is at SSD speed, capacity is enough for most usecases.
buxe2quec - Wednesday, July 24, 2019 - link
Something similar is called ZFS and caches on a SSD (L2ARC) the files you actually use, so that you can get 500 MB/s (or whatever your SSD does) even on big ones, without wasting SSD for the many but rarely used small ones.Not exactly what you asked, but AFAIK the best compromise so far.
igavus - Wednesday, July 24, 2019 - link
Also add compression to that and the effective read speeds are really magical with NVME ;-)CrazyElf - Wednesday, July 24, 2019 - link
Technically we have solid state hybrid drives (SSHD).tommo1982 - Sunday, July 28, 2019 - link
I wonder. How about moving whole MBR to MMR. It's nonvolatile, fast and you could save a lot on latency when heads don't have to seek where the data is stored.