Accelerate Your Hard Drive By Short Stroking
Table of contents
- 1. On The Stroke Of Performance: Hard Drive Short Stroking
- 2. Short Stroking: How It Works
- 3. Test Drives: Hitachi Ultrastar 15K450 and Deskstar 7K1000.B
- 4. “Niagara,” The Short Stroking Tool
- 5. Test Setup, Flash SSDs and Access Time
Flash SSDs dominate the news at present. If you believe the coverage from multiple Internet tech magazines, flash-based devices are about to take over the storage market. They’re perceived as being faster and more efficient than hard drives—and maybe a bit more expensive. The truth is different, though; flash memory still has only a tiny market share for many reasons. The good old hard drive isn’t dead, and we just found another reason for them to live on for a while: short stroking technology.
HDD Vs. SSD
I want to make very clear that we do not favor hard drives versus flash SSDs or vice versa. Each technology has its pros and cons, and each makes a lot of sense when deployed in a smart way. The advantages of SSDs are almost nonexistent access times and high I/O performance. In addition, flash memory has become extremely cheap for capacities of up to 32 GB. Finally, flash SSDs can be amazingly power efficient. Unfortunately, many flash SSD products on the market just are not there yet.
To understand these products, you need to know about the difference between single-level cell (SLC) and multi-level cell (MLC) flash. MLC allows for quick read operations, but the technology is not suitable for high-speed write operations, especially when they occur randomly. SLC flash is superior, but also more expensive. Some MLC flash SSDs are affordable, but the really juicy SLC flash SSDs are still very pricey.
Efficiency and reliability are additional issues. Performance and power efficiency are highly dependent on the quality of the controller used by a flash SSD product, as does power efficiency. There are some amazing products on the market, but they come at a price.
HDDs Holding Up
Conventional hard drives have become extremely affordable, with terabyte drives approaching—or even dropping below—the $100 mark. High capacities, low cost and ready availability clearly are the main reasons to keep using hard drives. And despite being based on magnetic, mechanic, and electronic components, HDDs are typically reliable and comparatively robust. In fact, enterprise class hard drives have to endure torture testing and demanding validation processes before they can be used for mission-critical applications. In contrast, the long term reliability of flash SSDs is still uncharted territory, and hence, they're avoided by many enterprise customers.
Fighting The Physical Disadvantage
The main disadvantage of conventional hard drives is the time delay that occurs each time the read/write heads are required to move. While this is not an issue for sequential read or write operations—where conventional hard drives are often still faster than most flash SSDs—it becomes a serious performance issue for random read and write operations.
Throughput of more than 150 MB/s for fast 15,000 RPM enterprise hard drives can decrease to only kilobytes per second once the hard drive has to access random information that may be distributed across the entire magnetic medium. In other words: the drive or the application loses a lot of time spent on head repositioning operations. What if this mechanical activity could be minimized? It can! We looked at an approach that is referred to as short stroking and tested a set of Hitachi hard drives that aim to reducing mechanical activity by utilizing only a fraction of the hard drives’ capacities.
- 1 / 2
- Next
-
Latest Internal Storage News
Latest Internal Storage reviews
- 16/05 – Intel SSD 330 Review: 60, 120, And 180 GB Models Benchmarked
- 11/05 – Do-It-Yourself: Upgrading Apple's 27" iMac With An SSD
- 01/05 – Marvell-Based SSDs From Corsair, Crucial, OCZ, And Plextor:...
- 26/04 – Best SSDs For The Money: April 2012
- 25/04 – Hitachi's 4 TB Hard Drives Take On The 3 TB Competition
Well we have been doing this for years. It is callad partitioning. You make one partition in the beginning of tthe drive to work faster for OS and software and leave rest of space for storage where You do not need speed.
It would be nice to see if partitioning leads to the same result, I suspect that it will, but would like to see the proof.
hmmm...
i dont know whether to discredit the benchmarks or not, but it would of seemed fairer to me if you had put the SSD drives into a RAID array to match the RAID array of the traditional HDDs. Now im unsure about this simply beacuse most SSD drives contain internal raid arrays yet they are still considered to be one drive.
can we have atleast one benchmark HDDx4 RAID 0 vs SSDx4 RAID 0
ainarssems - spot on! With regards to the SSDs in RAID0 - yes it would be nice to see that... but don't they scale pretty much linear?
Anyway - interesting article.
What about mini raid.
Mini Raid 1:
A 2/3/4 platter hard drive has a controller for each platter, and each platter is raided.
Mini Raid 2:
You can put 4 x 2.5/1.8 inch 1 platter drives in a 3.5 inch drive so you could raid those together.
Why has no manufacturer thought to make a dual or even a triple “arm” (seek/head arm) hard disk, I thought about this over 6 years ago and thought it was only a matter of time. Granted it wouldn’t probably be very energy efficient and the drive electronics would be very complex, but seek times would theoretically be reduced 3 fold; Thought-put (sequential) could also be increased by about the same margin.
That said it’s probably too late for it to be feasible especially as it seems we are heading for smaller and smaller eco drive/storage devices
The above is a good idea but I couldnt see it working well as it would require two or three seperate reads for one peice of data which would slow down the process even further.
Three platters on a drive already hold three seek arms, one for each platter. On a badly fragmented drive I would assume seek times are reduced drastically as it is!
Imagine the confusion with three seperate read/write heads on each arm!
why do they need arms? why don't they just get a new way of reading and writing information withou an arm, but with an object just like an lcd panel (with pixels) that could make an induction to the magnetic surface. Only needs to turn adress that location and begin transmiting bits, then to another place and transmit. No moving parts...
I think this might give HDD's some short term breathing space (3 years), then SSD's will just take over completely, the material costs, power consumption, space requirement and IO performance will just kick traditional HDD's out completely.
I think traditional magnetic media might have a place left in backup mediums but for any future performance benefit SSD's will be the only option.
Unless HDD' manufacturers have something up their sleeves?
I think this might give HDD's some short term breathing space (3 years), then SSD's will just take over completely, the material costs, power consumption, space requirement and IO performance will just kick traditional HDD's out completely.
I think traditional magnetic media might have a place left in backup mediums but for any future performance benefit SSD's will be the only option.
Unless HDD' manufacturers have something up their sleeves?
Well we have been doing this for years. It is callad partitioning. You make one partition in the beginning of tthe drive to work faster for OS and software and leave rest of space for storage where You do not need speed.
I don't agree with this statement, it might give you a benefit during start-up, but as soon as you access the data on the other partition the IO performance will degrade on the first partition!
Partitioning would only work if you leave the slow part of the drive unpartitioned.
I do believe this is a fair test though - even though the SSDs are not RAIDed, they probably cost as much as the entire array and controller, indicating that for the same outlay you get much much better performance from HDDs.
Wifiwolf: that description fits magnetic core storage, used in the 50s and 60s as a non-volatile RAM. Its modern equivalent is MRAM, which is a relatively new technology but looks quite promising as it's non-volatile like flash, but much faster (faster even than DRAM) and has unlimited write cycles. Unfortunately as far as I know it currently has no where near the density (and therefore low cost) of flash or DRAM.
My understanding is that partitioning a drive wont increase the access rate to better than the drive spec if you're still using the entire area of the drive e.g. if you have a consumer drive say a 7200rpm 500gb 16mb 8.5ms drive in your desktop at home partitioning wont increase the access rate more than the 8.5ms - you may get 8.4ms if you're really lucky but not better (unless you short stroke it). However not partitioning would mean you get considerably worse.
With shortstroking it is my understanding from this article that you can increase the access times considerably and looing at the results of the benchmarks you can do a lot better than flash drives in pretty much all areas except for web server with small bits of data from all over the place. A very informative article and something cost effective to consider for my next system.
I wish MS would apply this method standard in their defrag.
It'd be cool to have Windows and program files on the outside ring of a HD, while larger files (which often are movies or compressed archives) would be in the center.
It would definitely boost startup speed.
It would also make more sense to increase the numbers of heads on the arm.
That way one could effectively double the speed.
Instead of like on a CD or LP to have only one track,a HD could be dualtrack (2 spiral tracks per platter).
I always make my OS partition small with small clusters 4k and I also move stuff away from this drive andhave a second swap partition just afte the OS of 4GB on 2GB so OS is Usually no more than 8 GB ever butusually 4GB on a 640 GB drive and 4 of them RAID 0 this is 1GB on all of them does the head even move? other than to the next track. I had this discussion with maxtor 7 years ago and they used to publish the zone edges to avoid this slow down no one does this anymore
agree with prodigit80 it should be possible to configure a hard drive head system to work the way you want it to over zones and also if they added a second set of heads on the other side of the drive given the size of DVD drives etc this would actually double the speed of the drives and decrease access times by half without increasing power or spin speeds. Seagate didn't want to know the last time I asked them about this. I'd buy one though. Actually I'd probably buy 4 of them.
Actually hdd's with dual arms/heads have been made before as experiment but never made to mass production for some reason
Well you do not have to worry about HDD with multiple arms etc.
The next genration of HDD (call them as you want) they will store data in the mass of the platter and not on the surface.
That means in the space today you can store 1MB you will store 5 or 10 times more and the speed will be 5 and 10 times more respectivelly (the maximum output speeds). After couple of years they will manage to fit 50 or 60 times more capacity.
the same story with todays HDDs, few years ago we had 10GB per platter and now we have 500GB.
This technique I mentioned is real but we do not know yet when will be available to us. They need to experiment test it find the right material.
We may see those HDD's in 5 years (hopefully) or 10!!!
I believe we will see them when they will not manage to add more capacity in a platter and this limit is not far away trust me!!!