SAS vs VRaptor vs SSD

SSDs have existed since the 1970s..

Also only on a consumer level has it gone mainstream in the past few years. SSDs have been used for a long time.

http://www.ramsan.com/products/ramsan-5000.htm Enterprise/Datacenter level.

Hmm, I believe NAND flash wasn't invented till the 1980's, more specifically I think it was 1987. I agree SSDs have been used for a long time, longer than the 2 or so Joe is claiming.
 
If you do go with Vertex (either a single 60gb or 3x30GB RAID 0) you would be best off doing a clean install (and a lot of reading on the OCZ forum). This is a fun project, but there are a few things involved. You need to delete the partition that comes on the drive indisk managment, align your partition using diskpart (128k for single SSD), format using 4096 byte allocation unit size in disk management console, then do a clean install of vista onto the partition you created in your other OS without deleting it. Once you are all loaded up and happy, you need to disable defrag (super critical step) for that drive (and indexing too). If you don't do it this way you will have problems. There are also many other registry and services tweaks that will give you varying levels of results (like disabling superfetch etc).

RAID is even more complicated and there are lots of different combos of alignment, allocation unit, and block sizes to worry about, so read up first and you won't have to repeat. Also don't forget to flash to firmware 1199 first thing (assuming your new drive won't come with it) there is a huge performace increase in writes.

My Vertex Thread
http://www.hardforum.com/showthread.php?t=1401287

Or you can just buy an Intel X25-M, install windows and disable defrag and you'll be good to go. No need for all that extra crap- ;) That goes for almost any SLC drive as well.
 
Or you can just buy an Intel X25-M, install windows and disable defrag and you'll be good to go. No need for all that extra crap- ;) That goes for almost any SLC drive as well.
lol!

I did do a lot of work to get close to intel performance for $150 less... What is my research time worth?

I still say it was a fun and informative project. [H]
 
SSDs have existed since the 1970s..

Also only on a consumer level has it gone mainstream in the past few years. SSDs have been used for a long time.

http://www.ramsan.com/products/ramsan-5000.htm Enterprise/Datacenter level.

Today, when someone says "SSD" that implies and is well understood by anyone that knows a damned thing to mean Flash-RAM based storage, not actual Dynamic RAM powered storage. So... what I said was and still is accurate. SSD technology as it exists today has only been viable in terms of cost and manufacturing for roughly the past 2 calendar years and is quickly picking up steam (Moore's Law, anyone?).

Who knows where it will be 2 years from now and what kinds of densities we'll see, but I stand by my statement.

<besides, I've been using RAM "disks" since 1978 anyway... bleh> :D
 
well i have 4 15K.6 (the 450GB one) in my raid it is pretty fast, i just place an order with provantage for 6 of vertex 250GB (still back order) once i get that in hand i will compare and let you guys know then
 
The RAMSAN is not an SSD. Please stop calling it one. It's not even close. It's also considered by serious SSD people, to be nothing more than a joke.

It's software playing FC Target using a PCI bus on a basic PC platform, with a very large number of battery backed DIMMs. It does not use any sort of nonvolatile memory, only standard DRAM. Thusly, the RAMSAN is absolutely not an SSD by any definition of the term. Period.

Real SSD disks, and we're talking genuine enterprise dating back to about '95, come from a company called Curtis.
http://www.curtisssd.com/products/drives/
The key features that make these true enterprise:
64/8 ECC HAM internal storage mechanism, minimum 2 hour internal LiIon/LiPolymer battery backup, availability of internal mechanical backup.
And yes, they will blow the doors off any consumer SSD you throw them up against.

Touching on Fujitsu!
Don't. Sorry. They're out of the hard drive business. They make WONDERFUL drives, but will no more. I recommend against buying them. Hitachi GST's SAS drives are every bit their equal though. Don't touch Seagate - they've proven to be very slow in our production environment, to put it mildly. A 4+1 RAID5 of Seagate 15k SAS was slower than a 4+1 RAID5 of Hitachi 10k FC2.

For most loads, you will not see greater performance from SAS over SSD. You will over SATA without question. That's just a fact. However, you also need to question yourself on the cost. 10kRPM SAS will outperform SATA as well, and is significantly cheaper. My recommendation would be honestly, to go with whatever is cheapest, and acknowledge that realistically? You will not see enough performance difference during normal operation to justify the cost per gig I'm estimating here. 10k SAS in RAID1 or 4+1 RAID5 (NEVER have less than 5 disks in a RAID5 ever) will more than meet the needs of 99.5% of people.
 
I just snagged a 15k.5 73GB off of amazon for 50 dollars after I used my giftcard.

Gonna be on the prowl for another one, hopefully a used one that some bloke named wrong.
 
What are you using the drives for?

SSD's are the future, but after seeing the [H]ard video review of the Intel X25-M SSD, the future is not now. Especially for the price tag.

About to do a follow up as well. Will put the SAS into the mix. Got a couple of drives here that were middle-priced...
 
Hi Kyle,

Thanks for all of your work!

If you add Vertex to the mix, you may want to use the latest 1199 firmware since the performance differences are so dramatic on writes.

I'm pretty sure he's aware of the different firmwares... but maybe he should wait a few months till the 8th or 9th firmware comes out to be sure. :p
 
SSDs are not mature enough. Give 1 more year, but for now, I'd pick a Raptor.
 
Brahmzy,

Must have missed the threads that discussed this phantom 20k drive...thought it interesting, there was no follow up to the article I read. The original was published on June 6th of 2008, and it looks like 20 sites or so picked it up...but nothing since.

Wasn't aware it isn't in the pipeline...thanks for the heads up.

LC
 
I would go with SSDs just for the reduced heat and noise.

I've got a 64 GB SSD, 2 640 GB WD AAKS drives, and 3 older 15,000 RPM Seagate Cheetahs in my main rig. I would ditch the SCSI hardware if I could afford more SSDs.
 
I would go with SSDs just for the reduced heat and noise.

I've got a 64 GB SSD, 2 640 GB WD AAKS drives, and 3 older 15,000 RPM Seagate Cheetahs in my main rig. I would ditch the SCSI hardware if I could afford more SSDs.

Please don't compare ancient garbage Cheetahs to modern Fujitsu and Hitachi 15kRPM drives. They are night and day.
 
Please don't compare ancient garbage Cheetahs to modern Fujitsu and Hitachi 15kRPM drives. They are night and day.

They're not ancient, they're just not the latest model. Keep your garbage advice to yourself.

The mechanical to do 20kRPM is not even close, and WDC's mechanics typically lag two to four generations.

While I doubt there's a 20,000 RPM drive in the pipeline, do you have any proof that WD's tech "lags two to four generations" or did you pull that out of your ass?
 
For 15k SAS, varying widely by drive, with Fujitsu and Hitachi tied at first and Seagate dead last, your performance improvements on sequential are almost zero.

What do you use to compare the performance? Do you have a specific application that you've run these drives in to see this difference in performance? I am limited in experience when it comes to large RAID arrays with heavy IO requirements, but from what I've seen the Seagate's bench well, is that the only thing you believe they do well in? (Not being sarcastic, sincere questions- ;) )
 
While I doubt there's a 20,000 RPM drive in the pipeline, do you have any proof that WD's tech "lags two to four generations" or did you pull that out of your ass?

I call out of his ass.

Think about it, WD wouldn't be doing as well as they are if they were that far behind.

I mean come on... They do have the first 2TB drive. They do have the first consumer 2.5" 10k drive...

By your logic they would still have 2MB cache, be no bigger than 500GB, and would not even know how to make a 2.5" 10k drive.
 
They're not ancient, they're just not the latest model. Keep your garbage advice to yourself.



While I doubt there's a 20,000 RPM drive in the pipeline, do you have any proof that WD's tech "lags two to four generations" or did you pull that out of your ass?

You're making all these claims without even providing model numbers. I've looked at numbers and differences between 15k.5 and 15k.6 are pretty big (sometimes differences of 40MB/S).

If you're throwing in a 15k.3 or 15k.4 of course it's going to come in dead last.

Sounds like you don't know what you're talking about.

And I said I want some noise. A lot of people have been very helpful in this thread, but you're just rattling off some stuff without even reading my posts.
 
I call out of his ass.

Think about it, WD wouldn't be doing as well as they are if they were that far behind.

I mean come on... They do have the first 2TB drive. They do have the first consumer 2.5" 10k drive...

By your logic they would still have 2MB cache, be no bigger than 500GB, and would not even know how to make a 2.5" 10k drive.

And by your words, it's clear that you have no understanding of the definitions of the words "mechanical" or "typical." I'd refer you to a dictionary before opening your mouth.

Some of us are still amused by the WD Enterprise, I mean Raptors. Besides the fact that frankly, any idiot can put more cache or more platters into a drive. Quite literally, anyone here who could work an SMT soldering workstation could turn a Seagate with 16MB into a 64MB cache drive. All that means is they put in bigger SRAM and changed a few lines of code.
 
What do you use to compare the performance? Do you have a specific application that you've run these drives in to see this difference in performance? I am limited in experience when it comes to large RAID arrays with heavy IO requirements, but from what I've seen the Seagate's bench well, is that the only thing you believe they do well in? (Not being sarcastic, sincere questions- ;) )

Seagate's benchmarks are complete freaking crap. Large arrays are not where you use toys like HDTach or iozone. I only use the most brutal tests ever conceived by rational man. And by "brutal" I mean "I have made IBM PE's cry." And by "test machines" I mean "generally systems that nobody here can afford."

Brutal Test #1: Tivoli Storage Manager, over 40TB of Tivoli dedicated storage. (That's freaking huge, for those unfamiliar with it.)
Host systems can't be disclosed in detail - work's rule, not mine. I test with Suns that are smaller than E15k's and IBM POWER systems. I expect to be migrating production to a much larger system by June '10 (and screaming for faster disk as a result.) Tests involve selective backup of large filesystems, Oracle databases, and Exchange servers to disk. Then running from disk to tape, and tape to disk. Also dd throughput testing to evaluate LV layouts. (Restores and backups themselves are only good for testing network bandwidth.)

Brutal Test #2: Tivoli Storage Manager, only 2TB of direct-attach storage. But it's millions upon millions of small files.
Host systems are an IBM x3650 with dual quad-cores and 8GB, and an x3650 with single quad-core and 4GB. Host adapters are LSI 8480 and ServeRAID-10m (IIRC, may have wrong submodel.) Tests involve selective backups of large systems to disk, then running from disk to tape, and tape to disk. (Restores and backups themselves are only good for testing network bandwidth.)
In both cases, tape drives are LTO4. FC for Sun and POWER, SAS 3Gbit at one HBA per drive on xSeries.

Brutal Test #3: PeopleSoft 9.0 Financials and Supply Chain. If you do not have memory in multiples of 16GB, do not even try.
I can't disclose systems here. That's not my rule, that's work's rule. Sorry. What I can say is that this thrashes hard on random and IOPS; a single instance of the one test generates over 16,000 random I/Os per second. Do NOT attempt this with ordinary arrays, unless you want to sob. The other test is a heavy throughput and CPU load, using an archiving tool - I rely on the DBAs to give me data on this one. I don't know all the details. I do know it chews up disk.

All tests are run pretty much after going over every aspect of tuning and configuration with various experts. Then I tweak if they lied and it's unrealistic, or if I see it going faster with a different tuning. Monitoring is simple math (X MB in Y minutes == This Throughput,) IBM TotalStorage Productivity Center, Hyperic HQ, SNMP+MRTG on FC switches, and some undisclosed toys I occasionally get to borrow. Also, a large part of the data I have is obtained from regular monitoring of production loads. In other words, real world testing.

I've gotten my hands on directly, or in remote labs, basically everything from host-attached LSI 8480 through DS5k's and USP's. The bulk of my testing sits around midrange. I don't get to run all my tests, or necessarily any of my tests, on any or every given array. For example, I know roughly what TSM throughput would be on an IBM DS5300, but I haven't tested there. Part of it is simple mechanics - you cannot go >2Gbit/s read on a single 4Gbit FC interface. That doesn't mean a given array will do that, or even can, but it gives you your upper bound obviously.
Some of it also relies on extrapolation from a tested array or configuration. For example, if I were to test an array with 2Gbit ports that did 1Gbit/sec sequential read, and 1Gbit/sec sequential write, I know two things - one, the array is out of bandwidth. Two, locating two of these behind an IBM SVC is most likely going to give me a total of 2Gbit/sec sequential read, and 2Gbit/sec sequential write, when configured correctly. How? Because I know the SVC's limits and I know each array can do 1Gbit, which is well within the SVC's limits and capabilities. Now let's say that same array doing 1Gbit max has 4Gbit ports - I now know it's either A) out of internal loop bandwidth (2Gbit internal single loop?) or it's out of processing power at a nice round number. Either way, I know that there's an internal limitation rendering those fancy ports pretty much worthless.

Hopefully this answered the question. Like I said, there are some things I can't say for various reasons.
 
Who actually uses WD drives in an enterprise environment?

A good question. WD Enterprise actually refers to Western Digital's - at the time - "cutting edge" 10kRPM SCSI drives. Introduced around '96 or '97, and very very swiftly discontinued. The average lifespan, presuming you tried to keep the firmware at a "safe" level, was less than 12 months. I was saddled with a couple in RAID1 sets, and chewed up a drive every 4-5 months no matter how much cooling or babying I gave them. Always mechanical failure, always catastrophic. Extremely bad dampers, I suspect to this day. Either way, WD very quickly dropped them, because after about 9 months on the market, even resellers wouldn't touch them.
The Raptors are mechanically based on those drives.
 
A good question. WD Enterprise actually refers to Western Digital's - at the time - "cutting edge" 10kRPM SCSI drives. Introduced around '96 or '97, and very very swiftly discontinued. The average lifespan, presuming you tried to keep the firmware at a "safe" level, was less than 12 months. I was saddled with a couple in RAID1 sets, and chewed up a drive every 4-5 months no matter how much cooling or babying I gave them. Always mechanical failure, always catastrophic. Extremely bad dampers, I suspect to this day. Either way, WD very quickly dropped them, because after about 9 months on the market, even resellers wouldn't touch them.
The Raptors are mechanically based on those drives.

so their bad or good? O_O

For general use, such as just faster OS load up, opening programs/game loading, etc. Would you recommend a 300gb velociraptor?

EDIT: Just read storagereview.com, seems like a good drive to upgrade to :D
 
so their bad or good? O_O

For general use, such as just faster OS load up, opening programs/game loading, etc. Would you recommend a 300gb velociraptor?

EDIT: Just read storagereview.com, seems like a good drive to upgrade to :D

The time for SSD is here. I had 7 Velociraptors between 3 machines and am down to two now.

Go pick up a Vertex SSD or something if it's in your price range. Once you've used a single or preferably 2 or more in RAID0, there's just no going back.

For home PCs and laptops, mechanical drives are only for storage at this point, IMO.
 
so their bad or good? O_O

For general use, such as just faster OS load up, opening programs/game loading, etc. Would you recommend a 300gb velociraptor?

EDIT: Just read storagereview.com, seems like a good drive to upgrade to :D

Its certainly the best consumer drive available, although I have to say that the 640 GB Western Digital AAKS and AALS (Black Edition) drives are very fast and cost effective, especially if you put a couple in RAID 0. You can get two for ~$150. If absolute best performance is the priority I would get the V-Raptor, if you need more capacity I would go with the 640 GB drives.

If you just wanted a drive to install applications to I would get an SSD. They make good OS drives too, but you can run into stuttering issues with the current models.
 
And by your words, it's clear that you have no understanding of the definitions of the words "mechanical" or "typical." I'd refer you to a dictionary before opening your mouth.

Well technically I didn't say either of those words...

I'm just making the point WD isn't doing so bad, and that if their tech is so behind then why do we even bother with them? ANSWER THAT. Because you still aren't proving how they are behind so many generations, you are avoiding the initial question and bashing me for it.

Who actually uses WD drives in an enterprise environment?

IMO no one really, Raptors are called their enterprise drives but they are just fast consumer drives. If WD had SAS drives maybe...

At work we build small servers and sometimes use a Raptor as a boot drive per a customer request, but usually we don't use them unless in a gaming system because people know the name and it sells.
 
IMO no one really, Raptors are called their enterprise drives but they are just fast consumer drives. If WD had SAS drives maybe...

At work we build small servers and sometimes use a Raptor as a boot drive per a customer request, but usually we don't use them unless in a gaming system because people know the name and it sells.
The call their RE drives enterprise too, but apart from the TLER bit, they're identical to their desktop drives as far as I can tell.
 
The call their RE drives enterprise too, but apart from the TLER bit, they're identical to their desktop drives as far as I can tell.

Exactly and you can use the little utility floating on the web to make any WD drive act like a RE drive. Only difference then is the length of the warranty.
 
My experience with the Raptor drives has been fantastic. I'll say that of all the spindle types I've worked with over the years (somewhere on the order of 10,000+ spindles at this point from lowly workstations to multi-cabinet DMX4 with a load-out that starts where AreEss's systems stop) no drive as impressed me as much as the original 36GB Raptor. I have literally dozens of these (~ 70) as boot drives for design workstations, all in heavy use from the beginning, and I have not registered a single drive failure to date. I'm not as familiar with how the newer Raptor drives have held up, and I have been burned by WD's 10K SCSI attempts, but the original Raptor is among the most durable drives I've ever worked with.
 
You're making all these claims without even providing model numbers.
...
A lot of people have been very helpful in this thread, but you're just rattling off some stuff without even reading my posts.

I never made any performance claims or rattled off any numbers. All I said was that if I had the cash I would personally ditch my older SCSI/SAS hardware for SSDs rather than upgrade to newer SAS drives. I never claimed that the SAS drives would be slower, I said it was because I prefer the reduced heat and noise of the SSDs. I only mentioned it since SSDs had come up in the thread.

There was no need to get upset over it. I guess I'll refrain from offering my opinion to you in the future.
 
My experience with the Raptor drives has been fantastic. I'll say that of all the spindle types I've worked with over the years (somewhere on the order of 10,000+ spindles at this point from lowly workstations to multi-cabinet DMX4 with a load-out that starts where AreEss's systems stop) no drive as impressed me as much as the original 36GB Raptor. I have literally dozens of these (~ 70) as boot drives for design workstations, all in heavy use from the beginning, and I have not registered a single drive failure to date. I'm not as familiar with how the newer Raptor drives have held up, and I have been burned by WD's 10K SCSI attempts, but the original Raptor is among the most durable drives I've ever worked with.

The newer 150 was kind of a let down IMO, but the 36 and 74 have been very reliable.

Only time will tell for the VelociRaptors but I have seen a few dead ones at work right off the bat so, eh. :(

I never made any performance claims or rattled off any numbers. All I said was that if I had the cash I would personally ditch my older SCSI/SAS hardware for SSDs rather than upgrade to newer SAS drives. I never claimed that the SAS drives would be slower, I said it was because I prefer the reduced heat and noise of the SSDs. I only mentioned it since SSDs had come up in the thread.

There was no need to get upset over it. I guess I'll refrain from offering my opinion to you in the future.

I am looking to ditch my Raptors for the same reasons, less heat and noise.
 
Vraptors without the icepack, thats what i have right now.
Until the prices come down for SSD's im not buying.

I mean the performance is fine on even the low end SSD's but they are just overpriced.
 
My previous setup was 2 Savvio 15K RPM SAS drives in raid0. It was indeed very fast, w/ average access time of 5ms and sustained transfer rate of about 190-210MB/sec. But the rotational whine from the drives was driving me a bit nuts, so I swapped them out w/ SSDs. Currently I have two 64GB samsung SLCs in RAID0, and everything's working great so far. I don't really notice a significant performance boost versus the old SAS setup, but that near zero access time does help quite a bit for launching applications, bootup, etc. For example, launching the control panel, the icons load pretty much instantaneously, whereas previously you'd see the icons populate one by one, albeit quickly, due to the access time.

I haven't had personal experience w/ vraps, but I assume their performance would be similar to the savvios, and the firmware may be more optimized for desktop/workstation usage scenarios. You could probably get similar to SAS performance w/out buying SAS hardware by just settling with vraps.

For applications using really heavy loads, particularly with writes, I think SSDs may need to mature a bit and time-tested SAS drives might be more appropriate. Or SSDs might be more appropriate in a tiered storage using SAS drives as well. But for workstation use, SSDs are just fine and are much quieter to boot. I have my reservations about MLC, though, which is why I got SLC drives. If you can get SLCs for as cheap or cheaper than a vrap, I'd probably go with those over a vrap.

The samsungs I use listed for about 1000 last year, and now you can get them for around 200. That's less than a vrap, although you do get less storage space (64gb versus 300gb). It's not really a big deal for me since it's for the OS partition. I have a array of 1.5TB drives for bulk storage. This or next year might really be the year for SSDs to hit the mainstream, but I hope that some reasonably priced SLC offerings hit the mainstream. Seems like most manufacturers are focusing on the MLCs with funky firmware to try and hide the deficiencies of MLC. It kind of reminds me of the death of IPS and PVA in the mainstream display market.
 
Back
Top