It's down to $309.99 now. I'll be in a rather southern orbit in October, so maybe November will be a good time to buy.
SSDs - State of the Product?
- Thread starter ddrueding
- Start date
It's down to $309.99 now. I'll be in a rather southern orbit in October, so maybe November will be a good time to buy.
Pradeep
Storage? I am Storage!
http://www.semiaccurate.com/2009/09/22/larrabee-breaks-cover-last/
Scroll to the bottom:
"Intel seven SSD RAID on PCIe card"
Rated performance - 1.076 million IOPS
Scroll to the bottom:
"Intel seven SSD RAID on PCIe card"
Rated performance - 1.076 million IOPS
Stereodude
Not really a
You know the expression. If you have to ask...
Pradeep
Storage? I am Storage!
It was just shown at IDF. Given it's claimed performance 5 figures seems likely. If they are reliable they would be welcome in the datacenter space.
Pradeep
Storage? I am Storage!
Mercutio
Fatwah on Western Digital
There seems to be a LOT of FUD about using SSDs in the datacenter at this point. I know when I was looking a new server configs for my new standard platform last December that everything I read said to stay away from SSDs. I wonder how much of that is coming from "enterprise" drive vendors who are about to have their lunch eaten by SSDs vs. practical experience with SSDs in enterprise applications.
I really want to try to go that route at this point, if only because the low-volume systems I'm dealing with probably could see real world performance gains and I'm slightly skeptical of the claim that wear would really be an issue for the sorts of systems I'm dealing with.
Plus, an 80GB X25 is only about twice as expensive as a new 15k SCSI drive at this point. As an experiment it wouldn't be THAT expensive to try out.
I really want to try to go that route at this point, if only because the low-volume systems I'm dealing with probably could see real world performance gains and I'm slightly skeptical of the claim that wear would really be an issue for the sorts of systems I'm dealing with.
Plus, an 80GB X25 is only about twice as expensive as a new 15k SCSI drive at this point. As an experiment it wouldn't be THAT expensive to try out.
Pradeep
Storage? I am Storage!
For a standard array of drives with RAID 5 or 6, the X or E series seems OK. For something mission critical on a SAN? Well it's best not to roll your own in that environment. Just look at some of the firmware issues. Remember flash is a solid state device with a finite life, certain apps need indeterminate life. The true enterprise SSDs are massively over provisioned for this reason.
ddrueding
Fixture
I haven't run SSDs in RAID 5 or 6 yet (concerned the parity calcs will kill performance), but I have run them in 0, 1, and 10 and performance is fine. Once a 1TB array of SSDs gets cheap enough, I'll put them in a server under considerable load and see what happens
Fushigi
Storage Is My Life
In a database environment where you have 90+ percent reads and under 10% writes I think SSDs would be fine. That would include your ERPs, accounting systems, probably a lot of file servers, certainly web & app servers, and so on. But for truly dynamic data stores like, say, an Exchange server where there's lots of new and/or transitory data you'd be more likely to run into wear issues. Which is probably too bad since they'd benefit greatly from the high IOPS capability.
Our new enterprise storage management internally will tier storage based on speeds. We'll be offering a mixture of SATA, SAS, with both spindle and SSDs. The cost is very high for our internal SSD drives and obviously space is limited compared to spindle drives, but we can move customer data within the array to a different tier based on performance. Since our arrays work primarily by volatile memory cache, the I/O burden is there, but things are mirrored on the backend in case an SSD (or spindle drive) fails.
Hey guys, don't forget that many business applications are not drive constrained but rather network constrained: Adding an SSD in those will not help.
Fushigi is correct, there are the issues concerning an SSD's longevity. There are applications where there are lots of writes: Far more than personal use. SSD's longevity numbers are really based on normal writing for consumers, not continuous high volume writing that a business can generate. I'd worry about longevity too, if I were a business.
When running a business I'd be careful before swapping out HD's for SSD's but the potential benefit for some is definitely there.
Fushigi is correct, there are the issues concerning an SSD's longevity. There are applications where there are lots of writes: Far more than personal use. SSD's longevity numbers are really based on normal writing for consumers, not continuous high volume writing that a business can generate. I'd worry about longevity too, if I were a business.
When running a business I'd be careful before swapping out HD's for SSD's but the potential benefit for some is definitely there.
ddrueding
Fixture
An SSD failing is the same as a spindled drive failing. If you are a business and aren't prepared for that, you're screwed either way.
Mercutio
Fatwah on Western Digital
I still want to see some numbers. What does it take to kill an SSD under some crushing load? As far as I know, everything about the lifespan issues is theoretical. I've not seen anything about a real world high-volume setup.
Mercutio
Fatwah on Western Digital
I have a strong suspicion that there's some FUD coming from enterprise storage vendors.
Pradeep
Storage? I am Storage!
But who is the FUD directed at? AFAIK there are no "majors" using SATA "E" IBM drives in enterprise products. And the people throwing in a few drives for a build your own server aren't looking for an enterprise class solution.
When we see an E series drive with a SAS connector we could take it more seriously.
When we see an E series drive with a SAS connector we could take it more seriously.
Mercutio
Fatwah on Western Digital
What I'm primarily thinking about is the fact that you've got companies like EMC and NetApp building extremely high cost storage solutions based largely on dozens of spindles to achieve levels of performance, and several of the major players in spindle manufacture have no meaningful presence in the SSD market (WD bought in recently, but where are Seagate, Toshiba, Hitachi and Fujitsu?). If I can assemble a system that can compete on data throughput in a box that occupies 1/6 the rack space of a NetApp Flier, uses 1/3 the power and delivers equivalent performance at a similar cost, that's pretty compelling, and I wonder if it's in the interest of the guys who make such devices to keep us from looking at those sorts of solutions.
I'm notsuggesting that SSDs don't have a longevity issue, but I am saying that we REALLY don't know how bad it is, and I honestly suspect it might not be as bad as some are suggesting. In the meantime, we have these drives that aren't being marketed to desktop users because of cost and capacity and aren't being marketed to people who can afford them and could see massive benefits from using them because they don't have an SAS interface and may or may not last as long as the disk rotation on current SCSI/SAS drives.
Who exactly is supposed to be buying these drives right now? Just the elusive "power user?"
I'm notsuggesting that SSDs don't have a longevity issue, but I am saying that we REALLY don't know how bad it is, and I honestly suspect it might not be as bad as some are suggesting. In the meantime, we have these drives that aren't being marketed to desktop users because of cost and capacity and aren't being marketed to people who can afford them and could see massive benefits from using them because they don't have an SAS interface and may or may not last as long as the disk rotation on current SCSI/SAS drives.
Who exactly is supposed to be buying these drives right now? Just the elusive "power user?"
Well they might not have a real world clue on the drives longevity in deployment. All they have at the moment is data they gathered (hopefully gathered) from some production chips that they have hammered in the lab till it broke then used those results to calculate life based off of some "assumed" usage patterns. A company like Intel I can see doing testing like this but OCZ? I seriously have my doubts that they do anything more than slap supplier recommendations on there and ship.
Do they have anything explaining how the drive reports S.M.A.R.T. statistics?
Something else that I have been thinking about in relation to these drives is the sheer size of the reserved memory area on some of these things, especially considering the cost of it. So far we have a (good) explanation that they use it to alleviate some of the delay in this sector erase to reuse problem, but what if they are hedging lifetime bets and allocating plenty of space to take over for bad sectors? I haven't looked into this enough though and I don't know if percentage wise is this reserved space on par or way larger than what is reserved for a typical mechanical drive for bad sector allocation?
I am really waiting for someone like Anand, Kyle or maybe even Tom (if he can get out of the money vault) to set one of these things aside with a machine hammering rewrites 24/7 monitor the SMART info the drive is providing and see when something goes wrong with it.
Do they have anything explaining how the drive reports S.M.A.R.T. statistics?
Something else that I have been thinking about in relation to these drives is the sheer size of the reserved memory area on some of these things, especially considering the cost of it. So far we have a (good) explanation that they use it to alleviate some of the delay in this sector erase to reuse problem, but what if they are hedging lifetime bets and allocating plenty of space to take over for bad sectors? I haven't looked into this enough though and I don't know if percentage wise is this reserved space on par or way larger than what is reserved for a typical mechanical drive for bad sector allocation?
I am really waiting for someone like Anand, Kyle or maybe even Tom (if he can get out of the money vault) to set one of these things aside with a machine hammering rewrites 24/7 monitor the SMART info the drive is providing and see when something goes wrong with it.
ddrueding
Fixture
I remember reading that the X25-E had something like 50% extra set aside, the -M had 25%, and the OCZ drives 10%. As soon as the hot-swap adapters arrive, along with some Intel G2 SSDs from Buck, I'll beat the crap out of them on an ESXi machine hosting a dozen DB-intensive VMs until they die.
ddrueding
Fixture
Could it log SMART values as well?
In the GUI based VMware 1 there was an option to directly mount a disk that had some strong warnings against it. Now in the web based 2 I don't see that option anymore.
Does it need to be a VM? We might be able to scrape together an extra PC to do something like this. Myself I threw a load of crap away some time ago though.
You really want to potentially kill one of these just in the name of finding out lifespan? I mean it would be really great to find out and all but you are throwing some money away there.
ddrueding
Fixture
That is one of the reasons I'm more tempted to put it into production (albeit a heavy load), that way at least I get the benefit of it before it croaks. I was tempted to do it with an -E, -M and OCZ Vertex, but I'm getting cold feet.
timwhit
Hairy Aussie
timwhit
Hairy Aussie
No, still holding off a little while longer.
timwhit
Hairy Aussie
I finally pulled the trigger.
Stereodude
Not really a
Just FYI, your e-mail address is at the end of that URL.
That deal is rather tempting. The case looks pretty nice, and I wouldn't mind putting a SSD in my main desktop.
Ah crap, could an admin modify that URL for me? I don't mind you all having it but I can feel the spam coming in. EDIT: thanks for getting that so fast!
I bought one, I have been waiting and good timing for me as I am about to switch from win7 eval to retail as well. Seems you get two 3.5" converters in the deal as there is already one included with the drive, reviews on that freebie are mixed though so I will likely use the simple intel steel plate.