Skip to main content

ssd_vs_sata


Once upon a time, data storage performance (or I/O performance) was measured in "spindles":  The disk drive count in the array.  For those not intimately familiar with RAID (Redundant Array of Inexpensive Disks), this is the most common means of assembling multiple disk drives into one "logical" disk drive, so that data, such as a sizable file, may be stored in bits and pieces across them.  No single disk drive contains all of the file.  All the disk drives must be used together in tandem to access the full file, which is the whole point.

There are two reasons why RAID is used:  Redundancy (or data integrity/safety) and performance.  As the latter is sought after, the former becomes increasingly important. 

In most cases where RAID is used, it is performance that is actively sought after.  Even with the ever increasing spindle speeds (i.e.:  7.5k, to 10k, and now even 15k RPM speeds being common), this RPM is the Achilles Heel of disk drive performance.  Disk drives simply spend too much time -- basically their entire latency period -- waiting for the read/write needle (called a head) to physically move to the track being sought after on the disk drive, and there again waiting for the disk to physically rotate to the point where the specific part of the track (called a sector) is directly underneath the head.

You can only move a head so fast before the materials either warp or break apart, and you can only spin a disk so fast before the platter literally disintegrates and breaks apart. The industry is very limited by the physics at play on the materials being used.  

In the midst of this dilemma, enters a hard drive without the physics.  Or at least without the physics of inertia and momentum. Here enters the SSD: Solid State Disk (or Drive).  

Our hardware vendor brought these to our attention with our latest inquiry on pricing and availability of disk drives.  We were seeking those with the highest performance, but with the smallest capacity.  This is not an unusual request for us since our services are always pushing the performance envelope over the capacity envelope.  Unlike previous inquiries, however, our vendor came back with a suggestion that we try out the new SSD offering.

Now, SSD is nothing new.  To a limited degree, SSD is no different than a micro SD card or compact flash card used in phones and cameras.  For decades there have even been devices which permit the usage of RAM via a disk drive interface, some containing a battery back-up to retain data between power cycles.

However, the modern day SSD is an amalgamation of both.  Not nearly as speedy as outright RAM (SIMM, DIMM, SDRAM, DDR, or otherwise), but much more stable and non-volatile -- not requiring battery back-up.

The problem with the RAM drives of yester-year (data integrity issues aside) was cost.  One could purchase several piles of disk drives for the price in RAM of one modest sized disk drive.  Disk drive costs have largely remained the same over the years.  Naturally, capacities ever increase -- we started with 9GB drives whereas now the minimum available size is 73GB -- but the cost for each drive remains the same.  We don't need the capacity, we need the spindles and thereby the performance, so our costs have largely remained the same. We gained a lot by the increased spindle speeds, but not enough.  We have often employed dozens of disk drives to keep performance adequate for our more active sites, with unused free space reaching ludicrous proportions.  Meanwhile RAM (and related solid state technologies like usb thumb drives, SD cards, CF cards, etc.) have had explosive growth in capacity combined with plummeting costs to purchase.  Win-win.

So here we are now where it just so happens that SSDs are now fiscally viable for those seeking maximum performance with modest capacity.  A single pair of these drives should out-perform a fully loaded typical drive array (8-14 disks) in write throughput, with several times that in read performance.  And all for a cost less than that of a fully loaded array.

So, I sit here gazing at this 2.5" marvel realizing that unless I am way off my mark, this represents the most significant revolution in server technology in my more than 10 years in the field.  I think of the full racks of disk drive enclosures which could be replaced with a mere handfull of these SSDs -- never mind that double the number of these 2.5" form factor SSDs can fit in the same space as our standard 3.5" form factor disk drives.  It is only a matter of time before the whine of the disk drive is no more.  Some day, perhaps only a few years even, data centers are going to become much more silent, because servers are going to have screaming fast I/O, without the scream.   

Regards,
-Don

Attachments

Images (1)
  • ssd_vs_sata: SATA 3.5" vs SSD 2.5"

Add Comment

Comments (9)

Newest · Oldest · Popular
Greetings Steven:

Everywhere and anywhere that we would have been using 6-8+ disk arrays for performance -- you can kiss the notion of a JBOD good bye.  Our latest hardware addition, however, will be for some recent Enterprise customers.


Greetings Xochiquetzal:

Fantastic!  I enjoy explaining technology, but haven't had much opportunity since leaving college; used to drive a white board after classes -- frequently, others would join for a sort of informal brown bag (but without the lunch).  It was very satisfying, and helped to solidify and expand my understanding as well.

I appreciate your vote for the raise.  Social Strata (once Groupee, once Infopop) has always been very good to me.  For my part, I've been very grateful to be with such a visionary company.  I can't wait for the debut of some of the new community technologies coming up.  Very, very exciting stuff!


Regards,
-Don
DA
Think of RAID this way:

You have a 3 disk drive RAID array.  Using your vinyl record example, we'll consider each vinyl record on it's own LP player as a disk drive.

Now our records can only be played at single speed, i.e.: audio speed.  That's what we call the RPM, or the speed of the spindle.  This means for a 6 minute song, if we stored it on one of our records, it would take us 6 minutes to listen to the entire song.  It is not possible for us to play it any faster.  The limitation of our hardware only allows single speed.

Now, to understand the way RAID works, we have to first define a chunk size -- how much contiguous data we're going to keep on each member of the array.  Don't get lost here, we're going to call our chunk size "a lyric".  What this means, is that we'll store the 1st lyric of our 1st song on the 1st vinyl record.  The 2nd lyric of our 1st song will go on the 2nd vinyl record.  And, you guessed it, the 3rd lyric goes on the 3rd record.  For the 4th lyric, we have to start back over at the 1st record, of course.  And so on.

Now, when we want to listen to our 1st song, we can only listen to each record at single speed.  However, we can listen from all 3 records in tandem, playing the song pieces stored across all simultaneously.  An interesting aural experience, no doubt, but, then, you're a computer, so you don't care.  You just want the data, and you got it 2 minutes instead of 6.


Now to understand the difference between SSD and a typical disk drive, imagine you have that same 3 record array discussed above.  With a single SSD, we can read the same data not in 2 minutes, but in 3.6 seconds using just a single SSD.  In addition to this, our SSD has no motors, no delicate precision moving parts (the vinyl media, needle and needle arm), and takes up 1/2 the space of the record based system.  The SSD also costs marginally less for that single SSD than it does for the 3 record players (with records, of course).

The trade off, is that while you could easily store, say, 36 songs across all your record players, you can only store 2 or 3 songs on your SSD.

Now you can, of course, RAID multiple SSDs together to permit more data capacity (not to mention speed!!!), but doing so will increase the cost rapidly.  As such, SSDs are best leveraged in companies which push the performance envelope harder than capacity.


Let me know if that assists in understanding!


FYI:  The 3 disk RAID example provided above assumes RAID 0.  There are, unfortunately (for understanding) 4 commonly used categories of RAID:
  • 0: stripe -- preformance only (NO redundancy)
  • 1: mirror -- redundancy (some performance)
  • 6, 5, or 4: parity -- capacity & redundancy (good read performance, poor write performance)
  • 1+0 or 0+1: mirror w/stripe or vice versa -- massive performance & [massive w/1+0] redundancy (minimal capacity)

Shouldn't be hard to figure out which we use extensively, but that's neither here nor there for our example.  

Regards,
-Don
DA
Last edited by Don Aldrich (the donco)
Great blog Don... if I understood any of it!

But to make it something I might understand then.

A RAID is like an old vinyl record? You might only play one track a lot and whilst the part you play disintegrates, the rest of the record would perform fine if you could be bothered to move the tracks on? So your SSD drive spreads the wear and tear out across the disc without seemingly taking any time to reach or store stuff and thus making all tracks available at one click? Thereby extending the storage use?

Completely out of my depth here Don... but your enthusiasm is infectious!
G
Yes, for those who don't know, Don's been with us since the putting-together-furniture days.  We were discussing the new equipment in our Monday meeting, and I said, "we should blog about that!"  So we enticed Don out of the cave...er...his office and asked him to tell everyone about it.  Thanks for a great explanation, Don, feel free to blog any time!
Rosemary O'Neill
×
×
×
Link copied to your clipboard.
×