Mtron SSD 32 GB: Sweeping Performance with a Catch

[muk]

We predicted performance gains for Flash solid state drives. Mtron's 32 GB SSD proves that Flash drives can now beat the pants off mechanical hard drives, even though some rough edges still require smoothing.

http://www.tomshardware.com/2007/11/21/mtron_ssd_32_gb/index.html

 

[ethel]

One word sums it up really:

WOW!!

 

[rogerdandy]

"they're aren't always the best choice" <--- Direct Quote.

They're is a contraction of "They are". Aren't is not even a word. The proper choice would have been "They are not".

But seriously I can't wait for these drives to take over from my irritatingly loud mechanical HDDs.

 

[ghmage]

As the authors of this article mentioned the purchase price having the capacity to trigger a marriage crisis, I
couldn't help but remember back when a mid-life crisis related to a Harley. Still, as has been said in other threads
regarding solid state storage, the pricing is acceptable for enterprise usage.

It will be interesting to see the performance of storage increase to the point where it is no longer a bottleneck, at
least on a PC level.

 

[brundlefly76]

This review was way overdue from TomsHardware.

I have had this drive for 3 months!

MaximumPC reviewed these drives 4 MONTHS ago - since then several other sites have too, including Anandtech.

How does storage performance jump that high and Toms Hardware not report on it for 4 MONTHS?

I originally bought this drive to evaluate for a problem MySQL slow database query that we could just not solve in SQL. On a 15k Fujitsu MAS, the query took 40s - on the MTRON it took 4s.

However, it now lives in my laptop. When I got the drive, I was looking at upgrading my 14-mo old laptop (a 2GHz C2D Dell 1420). But once I put this drive in it, I realized that no new laptop would perform like this old laptop with the mtron. I used it to replace the new fastest laptop mechanical, and it was no contest. No heat, no noise, no delays, no nothing - launch 1 app or 4, this drive simply does not care. Battery lasts an extra hour (about 6 hours total). There is no heat and the fan never comes on anymore.

It is without question the best thing to happen to laptops since wireless, although I will TRY to wait for the price to come down and capacity go up for desktop use.

 

[joebob2000]

It can deliver up to 100 MB/s (95 MB/s on our reference test system) and at least two to three times better I/O performance than any mechanical hard drive, or sometimes many times more - even and especially including random writes, which is the achilles heel of Flash SSDs.

Is this what you meant to say? An 'achilles heel' is a weakness, and high speed random reads/writes are often touted as the main strength of any flash SSD.

 

[raptorxt]

"SanDisk was the first memory maker to provide a sample, and the Flash SSD SATA 5000 2.5" did well in our benchmarking suite. However, Flash does have a significant disadvantage: it's capabilities at random access are severly limited. This isn't much of a problem for regular users; they will still get a noticable performance boost out of a Flash SSD. However, frequent random access would still be executed faster by a fast, conventional hard drive than by a Flash SSD."

did i read that wrong or is it saying exactly what i think it is saying-
that if you use your computer a lot the random access time will "increase" significantly. I thought the random access time is the thing that made these drives so desirable because if the random access time "increases" significantly then my SAS drive destroys the SSD in every test.

please tell me if i am reading that statement totally wrong
dont have much experience in the solid state drives area of computing

 

[zenmaster]

Actually Random READs are the normal strength.
Writes of any kind are normally terrible.
He was pointing out this card did not have that problem.

 

[zenmaster]

I want to buy 5 of the 64gb models and put them into a RAID 5 so I have a good amount of disk space.

If anyone has a spare $15,000 that don't need........Just PM me :>

 

[Niva]

This is a good review, I wonder what's delaying Samsung with a wide release of their drives to the public, apparently you can only get one in a Dell. Only the old generation IDE drives can be purchased on newegg's website. Anyone have more info on this?

 

[dataland]

(Pingback) Many computer users know that their computer’s data is stored on a mechanical hard drive (aka: HDD, hard disk, or fixed disk drive). What many people don’t realize is just how old the technology is: The actual design consists of ... http://dataland.wordpress.com/2007/11/21/the-future-of-storage/

 

[ethel]

Aren't is not even a word

Yes it is - it's a perfectly acceptable and widely used contraction of 'are not'

 

[ewart]

Excellent I/O results, although am looking forward to faster transfer rates myself.

Just a note on comparing apples-with-apples (as much as that is possible in this field!). There is 'analysis' relating to mechanical drives not sustaining their transfer rates... well yep thats true but to be fair mechancial drives are an order of magnitude cheaper, and more to the point larger! What matters in a pure performance comparison in this setting is the first 32gb or 64gb of the drive, not the last 1gb on a 1000gb drive which is slow.

for example:

As the Seagate Barracuda 7200.10 or the Samsung Spinpoint F1 may reach better maximum results, these drives cannot sustain their high transfer rates.

■But they can sustain that speed over 32gb of the drive… e.g. looking at Toms recent SpinPoint F1 article, that hard drive reads and writes at over 100mb/s for about the first 500mb of the drive.. it averages 91.7 over the entire media and peaks at 118mb/s

However, the minimum performance still is 73.8 MB, while a Raptor will only write 43.2 MB/s as you get close to the 150 GB capacity.

■According to Toms's the average write transfer on the Samsung is 90mb/s over the entire 1000mb capacity.

A constant 94.5 MB/s read transfer rate is a new record for a Flash SSD. This is more than the WD Raptor can do.

■I'd agree it's more than western digitals 150gb aging raptor but less than their 750gb newer 7200rpm model.

So personally I'll put up with my crappy mechanical drives until these are affordable for end users. On the enterprise front, I'll install as many as possible

 

[nunyac]

Does anyone know how many times current flash memory cells can be written to before they becomes likely to fail? Early versions of flash memory technology were only rated for a thousand or so writes before the memory became unreliable.

 

[skittle]

Does anyone know how many times current flash memory cells can be written to before they becomes likely to fail? Early versions of flash memory technology were only rated for a thousand or so writes before the memory became unreliable.

you dont have to worry about memory going bad.. each sector has between 1million and 5million write cycles. that comes out to like 50-80 years before the memory starts to go bad. Im pretty sure this rivals the MTBF of mechanical drives these days.

 

[Crashman]

"they're aren't always the best choice" <--- Direct Quote.

They're is a contraction of "They are". Aren't is not even a word. The proper choice would have been "They are not".

But seriously I can't wait for these drives to take over from my irritatingly loud mechanical HDDs.

"They're not" and "They aren't" both work, so plz excuse the typo but there's no improper language in excess of that simple typo. Aren't is a word. Amn't isn't a word, which is why people use "Ain't".

Ain't is a properly formulated "new slang" for the non-existent "Amn't", but is easily replaced by the not-so-controversial "I'm not" rather than "I ain't".

Aren't has been around a long time and only the most arogant of sticklers would argue against it. This, as opposed to "Ain't" which most people argue invalidly against because it just sounds so coarse.

Good work Patrick, but please choose between "They aren't" and "They're not" simply to fix the typo so we don't have to see these fault-finding expiditions.

 

[natecar]

Write endurance is still a big issue with flash drives. Even at 1 million writes per sector (not per bit, but per 512 bytes, if I remember correctly), an active database can wear it out in a year or two. There are wear leveling algorithms, but totals writes can still overwhelm one of these drives. My comapny had an application that when we modeled it, was going to wear out a 5 million write-capable drive in 14 months. That's the average drive, so the 'less-than-average' one that fails first would have gone within the first year.
For many users, this is not a problem, but for the enterprise, be careful of where it goes. It is GREAT for reads, but the writes still can be an issue if you don't pay attention to them.

These SSDs have been around for a while, but are not mainstream yet. It will likely take the next round of technology (replacing current flash) before they become mainstream.

 

[bobwya]

Hi

My money is still on some kind of DDR-RAM drive (with ECC) like the HyperOS drive but built by someone that knows what they are doing!! I.e. that can implement a device with a 3Gb SATA-2 interface that can actually saturate the bus and use a mixture of DDR2 ECC RAM types.

Flash sounds great but the whole write wear cycle is a bit of an issue for me... Also the absolute transfer rates and somewhat limited (especially since they do a small latency to switch access to a different bank of memory).

Please note I am simply comparing SSD drives which demand a substantial price premium over regular mechanical drives!!

Are we going to see hybrid flash disks with say 32Mb+ cache to improve write performance at some point??

Bob

 

[ewart]

!! I.e. that can implement a device with a 3Gb SATA-2 interface that can actually saturate the bus and use a mixture of DDR2 ECC RAM types.

I'd like a decent DDR2 RAM based product on the market too.. and it's always nice to have a SATA-2 option, but I'd prefer PCIe to SATA-2 - if your going to go DRAM based why limit yourself to a pitiful 300mb/s.

Write endurance is still a big issue with flash drives. Even at 1 million writes per sector (not per bit, but per 512 bytes, if I remember correctly), an active database can wear it out in a year or two.

Yes I suspect such write endurance issues do apply to lower grade SLC based flash drives, or drives without strong wear levelling, and not industrial products such as the MTRON one reviewed here.. which I note has a industrial 5 year warrenty. I'm looking forward to a review of the fusionIO drive, which looks way better than MTRON. (noting that Toms reviewed the old MTRON series and the newer series (nov 07) now reads at 120mb/s.

Regardless, I find that with my clients active databases, we tend to upgrade the machine every year anyway. The less active databases are just fine on SAS.

 

[dfndoe]

Yes I suspect such write endurance issues do apply to lower grade SLC based flash drives, or drives without strong wear levelling, and not industrial products such as the MTRON one reviewed here.. which I note has a industrial 5 year warrenty. I'm looking forward to a review of the fusionIO drive, which looks way better than MTRON. (noting that Toms reviewed the old MTRON series and the newer series (nov 07) now reads at 120mb/s.

Regardless, I find that with my clients active databases, we tend to upgrade the machine every year anyway. The less active databases are just fine on SAS.

As someone who was bit on the ass by mtron SSD drives only a few months ago I'd like to point out that this claim is patently false. Out of 10 total 32gb mtron SSD drives we had a 30% failure rate when used as backing stores on a hi performance clustered mail server within the first 10 weeks. These drives are fast, but just like the competition, which I have also tested, they seem to have invested more money in developing fast throughput and less on a reliable controller, error checking and correction, spare block pooling, etc. I'm awaiting some eval units from STEC right now which I have high hopes for, after speaking with their implementation engineers on the phone and seeing the long list of DOD/NSA/DOT/MILspec certifications they have on data integrity (and in the case of their NSA spec drives, instant destroy... no joke) I suspect I'll find them up to the task. Seems they are having a hard time locating the eval stock for me to test their claims at the moment; although they were willing to ship me out a 40K usd fiberchannel drive, which I declined.

P.S. since rereading this post I sound like a S-TEC company shill/pr operative I feel it is necessary to point out that I have no relation to any of the companies which I've mentioned and have nothing to loose nor gain from sales or loss thereof from them. I'm only posting this to let people know of my personal experience.

P.P.S. mtron also recently re partnumbered their devices offering the drives which they considered "server appropriate" only 3 months back now as "desktop only" with a reduced warranty. Which leads me to believe they were willing to sell subpar equipment on the gamble that I wouldn't be able to wear them out so quick... bad gamble on their part.

Edit: Just looked at the pictures in the article again and noticed that the SN on the drive pictures is not even 300 units higher than the ones I had. A look back through my emails had the first two drives we received failing after 7 weeks with aprox 70gb of daily write operations and the third failed after 6 days at a similar load. Their own white paper and the engineers advised that I should have expected an MTBF of a million hours and a write life 120years with 50gb daily writes (that last bit is from memory and be a bit off.)

 

[ewart]

As someone who was bit on the ass by mtron SSD drives only a few months ago I'd like to point out that this claim is patently false.

The warrenty information comes from a PDF document entitled "Mtron line-up overview" dated 18-oct-2007 and emailed to me directly from Changcuk(David) Chung, who is the Managing Director / Sales & Marketing Department at mtron. If their product doesn't match their warrenty I can't imagine they will be in business long.. Let me cut and paste from the document here.

Mtron SSD PRO 7000 series
I/F: SATA 1.5Gb/s
F/F: Industrial Standard 2.5” & 3.5” form factor
Capacity: 16GB, 32GB, 64GB*
Casing: Aluminum
Performance:
Sustained Read: 120MB/s
Sustained Write: 90MB/s
Warranty: 5 years


Actually i see now a five year warrently it is listed on their web site even.

http://www.mtron.net/english/product/ProductDetail.asp?itemcode=MSP-SATA7025

Interesting to hear of your experience, hopefully you get your money back. Personally I don't think these drives are that fast.. I'm hoping for higher transfer rates from newer generation drives, esp considering seagates new mechanical SAS drive Cheetah 15K.6 is is expected to peak at about 164mb/s each (Q1 2008)... their current 15K.5 range peaks at 125mb/s...

cheers

 

[dfndoe]

The warrenty information comes from a PDF document entitled "Mtron line-up overview" dated 18-oct-2007 and emailed to me directly from Changcuk(David) Chung, who is the Managing Director / Sales & Marketing Department at mtron. If their product doesn't match their warrenty I can't imagine they will be in business long.. Let me cut and paste from the document here.

Mtron SSD PRO 7000 series
*snip*
cheers

Oh I already returned the drives through the vendor for 100% cash refund, I wasn't interested in warranty replacements when their failure rate was shown to be so high as to make them unreliable even in mirrored configurations. The warranty periods were changed recently and their models were renamed however their "PRO" line was added recently and the previous line was retargeted as a consumer grade product. However, looking at the picture it is clearly the same package right down to the serial number fitting into sequence (my vendor advised that the pro rated drives would be 'msp' series not 'msd); now perhaps they just went cheap with the markings and whatnot, but I wouldn't trust any thing important on one of those drives after my experience.

As to your being unimpressed with the speed, I too have 15k raptors in some of my servers and they do indeed rip for total throughput, but the SSDs (regardless of their flaws) totally tear them apart on random access loads such as (in my particular case) several million email operations a day per drive which have basically random access patterns and a mean file size of less than 10k.

 

[ewart]

If the failure rate is still the case one would hope they retract their 5 year warrenty on the industrial grade if it is not up to scratch.. time will tell.

sounds like the consumer grade definitely isn't up for any type of hammering.. pretty misleading of them to market it as suitable for servers.

Please post your evaluation results from alternate vendors when you have them, will be interesting to see how they performed. We too are holding out for decent I/O at a decent price.. I have been previously corresponding with David Flynn (CTO of fusionIO), and if that drive stacks up on all dimensions we be looking at that as a performance solution for several customers; fingers crossed.

 

[docimian]

http://www.dvnation.com/
has the 32 GB version for 725 dollars.
hopefully they can crack the 500 dollar mark soon

 

[dfndoe]

those 750 dollar 32gb ones are the "msd" series, not the "msp" professional series. I think it is pretty odd too that these mtrons drives are popping up all over the net now as "the fastest ssd" available when clearly their are faster options from manufacturers which don't target consumer class system builders. The Zeus IOPS SSD from s-tec was benchmarked over on IBM's site and the tester had to build a special testing rig with a fiberchannel switch as he was unable to saturate the drive with one machine doing i/o to it.

*I don't work for any of the entities mentioned in this post nor do I have any interest commercial or otherwise in sales of any devices or services mentioned.*