One of the main limitations I'm coming up against these days is write speed, especially when analysing a gig interface that's running at full chatter with a 1U rack mounted server.
We are having to build mini-arrays in these servers just to cope with the required write speed to keep up and not drop packets.
Would this technology improve write speeds or just capacity?
You're having trouble writing at GigE speeds? 125MB/s is not exactly superfast, I'm surprised that even a single spindle can't keep up with that.
Even so, if you need more speed, you need more spindles. I'd recommend something like ZFS with a 3 x 2 disc mirror (3 mirrors, striped), using 15k spindles, with a couple of mirrored SSDs as a cache device. Should get you close to around 10GigE speeds.
Lasers in HD's?
Why don't we skip the spinning/physical media altogether and go straight to laser/optical? Last I heard, it promised massive capacity, tiny latency, and with none of the limitations of magnetic media.
Hard drives with frickin' lasers?
Are they planning on turning them into gold next?
HAMR head with a frickin' laser
They must have had a geekgasm when they came up with that.
I sure hope they cost less than ONE MILLION DOLLARS.
I think you mean
HAMR head shark with a frikkin' laser.
Not exactly brand new tech...
Heat assisted recording tech has been used in magneto-optical storage before. I believe this is the tech used in Minidisc's, which was why they were rewritable (and later on, how they shoved 1GB of data onto 330MB disks with HiMD). In Minidisc drives the laser you saw at the bottom was not for burning to the disk, rather it just was for heating, it was the magnetic head above that did the actual writing.
Of course, this being applied to hard disks is new, and developing a head that has the required performance for disk speeds is no mean feat, so kudos to TDK for the work!
And in answer to Sir Runcible Spoon, it will improve performance, but as a side effect. The denser the platters are, the more bits will pass across the head per unit length. At the same speed, you will get more bits flying past the head for denser media, hence faster read/write speeds.
Now how much faster I'm not sure, hard disks are being specialised for large, slow storage. The "fast disk" area is increasingly being SSD based, with tiered storage systems. For example, I'm considering using http://bcache.evilpiepirate.org/ for ssd->HD caching of R/W IOPS.
With regards to read speed, I believe you're right that the increased density will cause an increase, so bonus! However, i believe i read somewhere that the additional time need to heat the media means that write speed will take a hit
"write speed will take a hit"
which is exactly what I was thinking, but without figures it's hard to gauge.
SSD's offer the speed, but not the capacity. We need to be able to dump several terabytes of data to disk at gigabit/s speeds.
^^this - bring back the MO drive
That's right the write method is exactly the same as the old magneto-optical drive. I suppose the difference is that (excluding that the drives are non-removable) the read method - which isn't mentioned - MO disks used the "Magneto-Optical Kerr Effect" which is an optical read - I wonder if they might try this too ...
No, this is not MO in either read or write. MO requires ferrimagnetic media produced with rare earth/ferromagnetic alloys with comp temps somewhere near room temperature. MO drives rely on a flood magnetic field which switches slowly to write along with the applied laser power. That means a two pass recording scheme where you must first pre-erase the media followed by a second pass to set the bits you want to be 1's. HAMR drives will still define the track with a near field optical spot, but the polarity of the bit will be set by the applied magnetic field in a single pass, i.e. high speed.
You are correct that MO uses an optical readback but you would never use it in a HAMR drive because your written spot is below the diffraction limit of any practical wavelength. This was a problem even with MO drives in the late 90's that the Japanese tried to get around using superresolution. I don't know if they ever shipped a product using that. Operating below the diffraction limit means that you would be reading back an evanescent wave which means low energy/poor SNR. Besides, the optics of a Kerr detector would be a pain in the ass and costly to integrate. A modern TMR reader is much more sensitive and that's exactly what will be used, or possibly a CPP-GMR reader if they ever work.
Spot heating is not a limit on recording speed or at least not a primary limit. It has some other implications. Data rate is not guaranteed to increase with increasing areal density. Doubling the number of tracks does nothing to the data rate of a given drive but still increases the capacity by a factor of 2.
..... HAMR time!
If the drive is rotating at the same rate 5k, 7.2k 10k RPM etc...
the bits are half the size (area), and not half the width.
the width determines how many you'll get one each track, so there'd be a speed increase of less than 50%.
Why do i have visions of users coming up to me shouting my laptops on fire, all I did was copy a 10GB file to my hard drive.
That's what I thought
What's the power consumption, and how hot will drives get when you're tring to integrate a new drive into an existing RAID?
Flame - says it all...
But will it get to market...
...before sold state storage overtakes it?
Enquiring gnomes wish to mine.
Everything better with LAZOR?
Actually not so much. With this, platters will be the new tape drive. A welded, gas-filled casing and a high-density medium with better-than-tape speed and hopefully better-than-tape shelf life.
However other technologies will control the bulk of live storage and transactions-on-the-fly. It's possible that we'll stop seeing platters on the shelf within five years, but it could easily be ten. For reasons of speed if nothing else, but capacity and durability have risen at a sharp pace.