LSI Implements SAS 12 Gb/s Interface

[btarunr]

LSI corporation is the first to implement new serial-attached SCSI (SAS) 12 Gb/s interface, geared for future generations of storage devices that can make use of that much bandwidth. For now, LSI proposes SAS expander chips that distribute that bandwidth among current-generation storage devices. The company displayed the world's first SATA 12 Gb/s add-on card, which uses PCI-Express 3.0 x8 interface to make sure there's enough system bus bandwidth. This card can connect to up to 44 SAS or SATA devices, and supports up to 2048 SAS addresses. It is backwards compatible with today's 6 Gb/s and 3 Gb/s devices.

By making use of the 12 Gb/s SAS Expander solution paired with 32 current-generation Seagate Savvio 15.3K RPM hard drives, LSI claims 58% increase in IOPS compared to a 6 Gb/s host controller, because of better bandwidth aggregation per drive. There's also a 68% increase in bandwidth yield. The array of 32 hard drives could dole out 3106.84 MB/s on IOMeter, and more significantly, over 1.01 million IOPS. As big as this number seems, it could be an IOMeter bug, because the numbers don't add up. Perhaps it's measuring IOPS from disk caches.



View at TechPowerUp Main Site

 

[n-ster]

over a million IOPS

 

[The Von Matrices]

over a million IOPS

Even more impressive considering it's being done on mechanical storage.

 

[W1zzard]

i think those IOPS are coming from cache.

IOPS for HDDs are calculated as 1/(Avg latency + Avg seek time). So: 1/(2 ms + 2.7 ms) = 213 IOPS

213 * 32 = 6,816 IOPS

i'd say user error using IOMeter

 

[John Doe]

HDD's aren't much use in datacenters anymore. An I/O drive would give superior I/O's in a much lower profile. Using a few PCI-E slots rather than a whole rack that needs power and cooling as well.

They should have done it on SLC SSD's to show all the capabilities of this controller (given it's application area)

still great stuff though.

 

[btarunr]

HDD's aren't much use in datacenters anymore.

They very much are. Single 250 GB SATA HDD is the most common storage option in leased servers. SSDs commonly start at an additional $50-odd per month for Intel SLC 50 GB.

 

[John Doe]

They very much are. Single 250 GB SATA HDD is the most common storage option in leased servers. SSDs commonly start at an additional $50-odd per month for Intel SLC 50 GB.

lol you bombed on this one. I was referring to enterprise datacenters. HDD's do nothing but suck power and underperform on an enterprise scale. Enterprise sized servers don't base their power on 250 GB HDD's. A single PCI-E or I/O SSD can replace tens, if not a few hundreds of them. Look up "ZeusIOPS".

 

[btarunr]

lol you bombed on this one. I was referring to enterprise datacenters. HDD's do nothing but suck power and underperform on an enterprise scale. Enterprise sized servers don't base their power on 250 GB HDD's. A single PCI-E or I/O SSD can replace tens, if not a few hundreds of them. Look up "ZeusIOPS".

No yuo bombed on this one. SSDs are far too unreliable and have far too high price per GB to "replace" HDDs in enterprise datacenters. By these, yes, I mean "enterprise-grade" SSDs. No enterprise with half a brain replaces most of its HDDs with SSDs are you're suggesting. Instead, they use SSDs only to temporarily hold the "hot" parts of their databases, and constantly keep them in sync with hard drives arrays, which are infinitely more in number.

 

[John Doe]

No yuo bombed on this one. SSDs are far too unreliable to "replace" HDDs in enterprise datacenters. They use SSDs only to hold the "hot" parts of their databases, and quickly archive to hard drives, which are infinitely more in number.

Uhm, yeah. Time to do some research on what I just mentioned. Look it up then tell me what the IBM guys say about it. SSD reliability (especially real SLC like ZeusIOP's, single cell only) is much more improved. It's more reliable than a convertional HDD. The SLC you see on the market (like Intel X25 Extreme) still use a few cells to store the data per-cell. They don't compare to the reliability of ZeusIOPS. The ZeusIOPS is an OEM only drive that enterprise business use instead of hundreds of HDD's. The memory is uses is like ECC RAM (as in reliability). It costs thousands of Euro's per drive and can only be ordered from STEC themselves.

 

[btarunr]

Uhm, yeah. Time to do some research on what I just mentioned. Look it up then tell me what the IBM guys say about it. SSD reliability (especially real SLC like ZeusIOP's, single cell only) is much more improved. It's more reliable than a convertional HDD. The SLC you see on the market (like Intel X25 Extreme) still use a few cells to store the data per-cell. They don't compare to the reliability of ZeusIOPS. The ZeusIOPS is an OEM only drive that enterprise business use instead of hundreds of HDD's. The memory is uses is like ECC RAM (as in reliability). It costs thousands of Euro's per drive and can only be ordered from STEC themselves.

Again, no. What IBM thinks doesn't reflect on how today's enterprises are storing data. And no, not even SLC SSDs are more reliable than enterprise hard drives. IBM is merely endorsing SSDs in its ZeusIOPS paper, it's in no way showing the spread of today's enterprise data.

And if we're talking research: http://www.snia.org/sites/default/files/AnilVasudeva_Are_SSDs_Ready_Enterprise_Storage_Systemsv4.pdf

^look at how that confirms my views on the spread of enterprise data.

 

[John Doe]

Talking about researches, you don't know what you're talking about. So I suggest you to do some more on ZeusIOPS. Especially read what storage professionals say about it. It's the future of store. It can reproduce the I/O's of hundreds of drives by itself alone. It's extremely reliable can easily, constantly be backed-up. You can use a few consulting each other for the potential of a thousand drives. It's ridicilious to use 1000 drives instead of 5 ZeusIOPS with backup. Do you have an idea on how much it takes to cool and power 1000 drives?

That research is showing HDD's in use because only the rich and big business can afford such setup. Companies are still on HDD's because of their price and availability. Such HW isn't sold to every server. ZeusIOPS is in no way a traditional SSD, and I/O drives will take over. But it's a slow transaction at this rate.

Seriously, inform yourself from the info out here.

 

[btarunr]

Talking about researches, you don't know what you're talking about.

Go to bed.

1a

1b

2a

2b

Seriously, go to bed. At this moment you are simply unable to debate.

 

[John Doe]

Go to bed.

Seriously, go to bed. At this moment you are simply unable to debate.

Yes, I'm. Re-read my post. They aren't in use because they aren't available options to all servers. Only the richer and more modern servers have moved up to those. And the thing is, they WILL be used in the future. We won't have HDD's by the time all flash become as reliable and as fast as ZeusIOPS.

Companies that're building from the ground up are moving towards enterprise SLC SSD's. No-one in his right mind would get a load of HDD's over SSD's at this time and date.

 

[Zakin]

Out of curiosity John, are you like paid by IBM to endorse them?

 

[John Doe]

Out of curiosity John, are you like paid by IBM to endorse them?

That's a pretty absurdist statement.

But yeah, they used to pay me... to talk crap about eVGA on their forums... to get myself banned.

http://www.evga.com/forums/fb.ashx?m=1127027

 

[Zakin]

That topic you had entirely valid points. As for this one and many others I have to agree with BTA, where as Zeus is a great technology for one the Enterprise even if they wanted it I'm sure are still looking at plans to overhaul to go to it. As I firmly also believe most of the enterprise is still running off of mechanical. I also don't even remotely see how my statement was absurd, I actually was convinced you were trolling this entire topic.

 

[John Doe]

That topic you had entirely valid points. As for this one and many others I have to agree with BTA, where as Zeus is a great technology for one the Enterprise even if they wanted it I'm sure are still looking at plans to overhaul to go to it. As I firmly also believe most of the enterprise is still running off of mechanical. I also don't even remotely see how my statement was absurd, I actually was convinced you were trolling this entire topic.

I'm not, why should I? You didn't need to nitpick on my post right in the first post. Enterprise is running mechanical but that's going to change. My original post stated;

"HDD's aren't much use in datacenters anymore".

I didn't say they aren't used anymore. I tried to say it's more logical to invest in SSD's rather than to buy hundreds of drives for this kind of application (Input/Output's). It's much easier to do it that way. See the original post? The topic is about the amount of IOPS LSI came up with on their controller.

 

[Zakin]

Right I understood that, although I also stated that I agree with BTA, that was also half of his argument that the industry also was not running SSDs. I personally even with Zeus wouldn't be surprised if industries would not be attracted to them. The reliability of SSDs is still a toss up just because of history, until the big wigs in the corporations realize that it's worthwhile then they will jump on the bandwagon. I actually highly doubt they will invest now though, and with companies continuing to put out new technology that supports the mechanical more and more they are also pushed into not seeing a point to change how they go about it.

 

[John Doe]

And that's the problem. People are still thinking HDD's are the only solid option. They aren't. For a fact, current MLC has much writes that you don't have to worry about degradition like people did a few years ago (for home usage).

That example aside, the logic used in flash like Zeus is perfect. It's great when you can reapplicate I/O levels of that much drives under just one 120 GB SSD. It really is the way to go. If you read what some storage experts say, they repeat the things I said in here.

People don't bother and are being stubborn on SAS/SCSI drives. They don't realize non-volatile flash has come a long way and will take over. Now the question is when.

 

[btarunr]

"HDD's aren't much use in datacenters anymore".

You highlighted the wrong part of your statement. You said:

"HDD's aren't much use in datacenters anymore".

That statement is incorrect, I've provided statistics to refute that. Mechanical drives are far from being "aren't much use in datacenters anymore."

 

[John Doe]

You highlighted the wrong part of your statement. You said:

"HDD's aren't much use in datacenters anymore". The most logical interpretation of that statement is that you claim HDDs.

That statement is incorrect, I've provided statistics to refute that. Mechanical drives are far from being "aren't much use in datacenters anymore"

Okay sir, let me clear up further then.

HDD's aren't much use in continously accessed data. That is high I/O's, which is what the OP is on about.

For storage though, yeah, they are. HDD's and especially tape for mass storage.

 

[NdMk2o1o]

Okay sir, let me clear up further than.

HDD's aren't much use in continously accessed data. That is high I/O's, which is what the OP is on about.

For storage though, yeah, they are. HDD's and especially Tape for mass storage.

They might not be of "much use" however even in enterprise datacentres they are still the most common form of storage/data by a country mile compared to SSD/IO drives

 

[Easy Rhino]

that was a good read. anyone who thinks HDDs aren't much use in enterprise datacenters today is clearly on drugs. i'm glad i didn't get sucked in that nonsense.

 

[repman244]

People don't bother and are being stubborn on SAS/SCSI drives. They don't realize non-volatile flash has come a long way and will take over. Now the question is when.

Yes SSD will take over, but right now? Not even close, SSD's in enterprise solutions is still quite rare, it is getting in there but SAS HDD's still hold their ground.
The biggest issue with SSD is their number of writes, they would be raped in heavy write environment. And this I say from my own experience.
SCSI/SAS HDD's have no issues working for 5 years 24/7 while getting hammered, now let me see an SSD do that.

 

[Disparia]

Well if you believe post #21 (not saying anyone should) and he only meant high I/O scenarios this whole time, then he does have at least a little standing. Million IOPS systems were comprised of ~750 short stroked 15K hard drives only five years ago. Performance of two full racks now can be had in 4U thanks to SSDs.