SAS hits 12Gbit/sec, when will SATA match that?

[VirtualLarry]

http://www.tweaktown.com/news/25459...ds_first_12gb_s_sas_enterprise_ssd/index.html

 

[Cerb]

I think it depends on how well the whole PCI-e integration thing will work out. That works out well for consumer gear, then maybe. Otherwise, definitely.

SATA's reasonable max has yet been determined. IIRC, the maximum will have been reached when it needs a <1m cable.

 

[ShintaiDK]

SAS is usually also shared.

SATA12 would only yield very deminishing returns on the desktop for now.

 

[rsutoratosu]

What do you mean by shared ?

http://en.wikipedia.org/wiki/Serial_attached_SCSI

Mentions nothing about shared on the wiki. Only legacy scsi is shared. I understand device to controller can be dedicated but after the data travels from controller to other parts of the processor, those lane could be shared.

 

[Soulkeeper]

maybe we see sata4 for the next wave of chipsets/platforms on shelves early 2013 maybe ?

 

[Hellhammer]

maybe we see sata4 for the next wave of chipsets/platforms on shelves early 2013 maybe ?

Lynx Point (i.e. Intel 8-series chipsets) is still SATA 6Gbps.

 

[Smoblikat]

If youre needing THAT much bandwith you should just use PCI-e, no HDD can even saturate sata 2 anyway.

 

[Cerb]

What do you mean by shared ?

http://en.wikipedia.org/wiki/Serial_attached_SCSI

Mentions nothing about shared on the wiki. Only legacy scsi is shared. I understand device to controller can be dedicated but after the data travels from controller to other parts of the processor, those lane could be shared.

Read it again, without skimming. SAS controllers and drives typically support SATA-compatible port multipliers, and SAS' own expander standard.

 

[imagoon]

What do you mean by shared ?

http://en.wikipedia.org/wiki/Serial_attached_SCSI

Mentions nothing about shared on the wiki. Only legacy scsi is shared. I understand device to controller can be dedicated but after the data travels from controller to other parts of the processor, those lane could be shared.

SAS is a point to multipoint protocol. Channels can be expanded and support many disks on one channel. This is how a single 4 channel SFF-8088 cable can operate a 20 disk DAS array. Add on to that, most SAS HBAs are dual ported.

 

[tweakboy]

sata 3.0 6gbps will be the connection on Haswell . We are not going to see 12gbps desktops until a few years , maybe 3 to 4 years. Look how long USB 2.0 lasted,, same thing...

 

[mv2devnull]

The 12gbps SAS standard is apparently not final yet. Nevertheless, SSD makers showcase examples. I'd say you will be able to get a PCIe SAS card to your desktop before you can buy a server with integrated 12gpbs SAS chip and a 12gbps SATA integrated into Intel's chip obviously takes a lot longer.

As said above, a PCIe SSD makes more sense than SAS/SATA controller + drive on a desktop. Server-market is different, however. You want to hotplug storage to your blades and SAS/SATA supports that nicely.

 

[Douglaster]

LOL, wanna see a cost-effective SAS board first.

 

[vss1980]

I think the thing to keep in mind here is the additional flexibility of SAS and how it might be used with switches / multipliers.

I suppose in a scenario where you had 4 x 3 Gigabit drives on a SAS switch/multiplier you'd need the 12 Gb/sec link to keep them all fed, assuming they could all saturate the bus.....

Even current SSD drives are a way off saturating a 12Gb/sec link.... but hey at least it gives the manufacturers something to aim at

 

[Cerb]

LOL, wanna see a cost-effective SAS board first.

From a consumer perspective, you won't. Being able to get higher profits by making the server parts use a different communications standard is basically the reason SAS exists as a semi-separate entity.

 

[imagoon]

LOL, wanna see a cost-effective SAS board first.

Perc6i / LSI / Promise / Adaptec make quite a few cheap ones. My Perc6i was $80 off ebay with BBU.

 

[imagoon]

From a consumer perspective, you won't. Being able to get higher profits by making the server parts use a different communications standard is basically the reason SAS exists as a semi-separate entity.

Uh. No. SAS exists because the needs of the server world isn't matched by the consumer desktop.

 

[vss1980]

From a consumer perspective, you won't. Being able to get higher profits by making the server parts use a different communications standard is basically the reason SAS exists as a semi-separate entity.

Not to mention the significantly increased core logic requirements such controllers usually have onboard to help push the extra performance, as well as allowing for features in some cases like battery backed up cache ram, etc.
SCSI and SAS as its successor will always have far more complicated HBA controllers than a normal PATA/SATA controller.

I still remember the days of moving away from PIO IDE controllers bogging down the CPU and the difference that made.... in a way, apart from maybe the addition of NCQ to offload even more work to the controllers/drives themselves, not much has changed since then in the consumer space.

 

[Cerb]

Uh. No. SAS exists because the needs of the server world isn't matched by the consumer desktop.

You say no, but then you aren't disagreeing at all, with what you said after that.

If everything were just SAS, with a few less-used features being optional, it would be just as good, but wouldn't command the price premiums it does, which are mostly due to being used in a smaller market that accepts higher margin. Additional logic needed isn't enough to matter, in the long run--it just gets cheaper over time--and the interoperability would allow cheaper hardware to be used with expensive hardware, which only grudgingly can happen today (SATA drives on SAS controllers).

It is not in the interests of any storage company to allow SAS market segments to get anywhere near as cheap as mainstream parts, and any business that needs faster and/or more reliable storage is willing to pay more for it. The added costs in development and testing would be minimal, and go down to negligible after the 2nd or 3rd generation of complete integration. But since it would add little to nothing to most desktop users, it's better to keep it separate, allowing higher margins to stay higher, so long as those margins more than make up for the costs of making it.

There's nothing morally wrong with such behavior, since businesses that care are going to have to pay more anyway, to some degree, and developing faster and/or more reliable products costs more, and the average consumer won't pay the extra for that, but the SCSI line in the sand is all about servers being higher margin, and keeping it that way.

 

[vss1980]

There is one other aspect that is missed out as part of the cost....

Warranty and support. Like it or not, you are paying a premium in the corporate/enterprise market for the expectation that it works and if it doesn't that you get a professional response. From experience, you get a definite difference in support from companies depending which sector the products are from and the support expected.

Additionally you also have potential 'business assurance' costs added in where they will stockpile certain products for potential support usage later after the production run has ceased - the manufacturers are going to make certain their costs are covered somehow.

It's why Intel, for example, charge so much for motherboards or SSDs or network cards (regardless of features included or not compared to the competition).

 

[bunnyfubbles]

meh, the new intel chipsets should finally allow for more than 2 x SATA6, and I'd be just as happy to keep adding SSDs to a RAID-0 array in order to achieve faster speeds and more capacity (especially now that Intel is supporting TRIM over RAID-0 for their newest chipsets), I'm not really anxious for faster individual drives when SSDs are still sparse enough in terms of capacity to where RAID-0 makes sense if only for increasing capacity let alone with getting a pretty substantial boost in speed.

 

[imagoon]

You say no, but then you aren't disagreeing at all, with what you said after that.

If everything were just SAS, with a few less-used features being optional, it would be just as good, but wouldn't command the price premiums it does, which are mostly due to being used in a smaller market that accepts higher margin. Additional logic needed isn't enough to matter, in the long run--it just gets cheaper over time--and the interoperability would allow cheaper hardware to be used with expensive hardware, which only grudgingly can happen today (SATA drives on SAS controllers).

It is not in the interests of any storage company to allow SAS market segments to get anywhere near as cheap as mainstream parts, and any business that needs faster and/or more reliable storage is willing to pay more for it. The added costs in development and testing would be minimal, and go down to negligible after the 2nd or 3rd generation of complete integration. But since it would add little to nothing to most desktop users, it's better to keep it separate, allowing higher margins to stay higher, so long as those margins more than make up for the costs of making it.

There's nothing morally wrong with such behavior, since businesses that care are going to have to pay more anyway, to some degree, and developing faster and/or more reliable products costs more, and the average consumer won't pay the extra for that, but the SCSI line in the sand is all about servers being higher margin, and keeping it that way.

Again. No.

SAS is Serial Attached SCSI. These devices have micros on them that have minimal or no place in the home consumer environment. End users do not need what amounts to a networking protocol where multiple devices communicate with the central controller and each other. The do not need the multipath connections. SAS provides two connections to allow for redundant access to the disks which is managed by the on HDD controller and the SAS roots. The cards often have battery backed up RAM and multiple IO (xor) engine chips to handle the common RAID and disk management tasks. Add in longer warranties and normally much more advanced HDA's, there is significant cost delta between the consumer SATA and SAS. This is just a couple of things that all add cost. Storage is cut throat in the server market. If someone could cut the disk costs by 10% they would do it in a heart beat because it would translate to a huge volume for them.

 

[Cerb]

Aside from warranties, which aren't really different in length (1,3, and 5 years are common), what of those other features would have been supported on consumer drives and controllers? Pretty much none, much like the old SCSI days when workstations would have controllers without nearly the capabilities of the big bad server cards.

 

[imagoon]

Aside from warranties, which aren't really different in length (1,3, and 5 years are common), what of those other features would have been supported on consumer drives and controllers? Pretty much none, much like the old SCSI days when workstations would have controllers without nearly the capabilities of the big bad server cards.

I assume you are replying to me. Your question makes no sense. If you like to buy a multi-controller 15k SCSI drive and plug it in to an ISA SCSI card... More power to you.

 

[Emulex]

SATA is not LVD, its signalling is far inferior so it won't scale as well. They need to ditch sata and just move to sas. LVD was introduced what back in the ultra-160/320 days? before that you had HVD which could run 30-40 feet scsi.

Sata signalling has to grow up to go faster. And 12GB of half-duplex doesn't scale either.

I'm really surprised that we haven't reached a point where motherboards just have both SAS and SATA these days. IIRC all the new workstation/server boards have sas controllers on them.

 

[VirtualLarry]

I'm really surprised that we haven't reached a point where motherboards just have both SAS and SATA these days. IIRC all the new workstation/server boards have sas controllers on them.

My MSI K9A2 Platinum mobo has a 4-port Promise SAS controller integrated into it.