Jump to content

Talk:Serial Attached SCSI

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Q + A

[edit]

Q: It would be interesting to see even a short list of points about how SAS compares to SATA - what limitations each one has the the other doesn't?

A: As noted in the article, SATA is limited to direct-attached connection of ATA drives only and does not support SAS drives. SAS supports SATA drives (in fact, it uses the same internal narrow cables), a complex topology, a SCSI command set, and for SAS-native drives, a globally unique drive identifier. 70.162.144.181 02:46, 6 October 2006 (UTC) A: Sata does actually support port multipliers but only one level of them and only one initiator and relatively few ports. SAS allows for multiple initiators and networks of expanders. 21:46, 13 April 2008 (UTC)[reply]

Q: I'd like to see something about performance of SAS relative to SCSI or SATA or IDE. Probably best as a link out than directly here. Leodirac 23:46, 5 January 2007 (UTC)[reply]

Q: It would be nice if SAS Expanders/Fanouts were explained more thoroughly. IE What the devices look like and how they interact with the controller. More of a broad explanation and implementation rather then ultra technical. Diagram, pictures, etc.

Q: The article says that SAS is backward compatible with SATA. What does that mean exactly? A: Sata drives can be connected to a SAS system but not vice-versa. Plugwash (talk) 21:46, 13 April 2008 (UTC)[reply]

Q: I would like to see a brief History of SAS in the document -- when was it introduced to general public? Harry Pehkonen (talk) 15:21, 26 October 2011 (UTC)[reply]

Q: Electrically SAS is a port but not a bus. Did I miss smthng? — Preceding unsigned comment added by 185.57.28.108 (talk) 14:09, 1 April 2017 (UTC)[reply]

SAS vs parallel SCSI

[edit]

Since parallel SCSI also support hot swap. I've remove the following line

--Sltan 05:48, 23 January 2007 (UTC)[reply]

Parallel SCSI does not really support hot swapping..You an attach / unattach devices without powering down the computer, but you need to manually do a bus-rescan every single time the bus is changed. If you do not do a bus rescan, the changes will not show.

With SAS, this is automatic because it is a point to point serial link, not a bus. 208.69.85.39 (talk) 21:19, 8 December 2008 (UTC)surge[reply]

-- comments by ard >: Hotswap with parallel scsi comes from the enclose controller. That one knows when a new disk is inserted, and hence a bus scan can be targeted to one id, and initiated automatically. Sata/Sas practice the same thing except that the hot-swap is speced as part of the bus, and hence is supported in the controller, and hence the *driver* knows when to scan instead of some enclosure daemon that polls the enclosure on events. Furthermore: currently it says that parallel scsi is multidrop, but it's just a bus which can host multiple initiators and targets. Initiators can be targets as well (that has been supported in some of the better operating systems). —Preceding unsigned comment added by 217.196.45.116 (talk) 11:34, 27 August 2010 (UTC)[reply]

Cleanup

[edit]

There are references for this article but they do not follow Wikipedia's guideline for citing sources at WP:CITE. Ceros 04:15, 5 December 2006 (UTC)[reply]

SATA 1 and SATA 2?

[edit]

In the article about SATA it can be read that SATA II shouldn't be confused with SATA/300. Which is said on www.sata-io.org (the ones who make the standard). So the comparsion between SATA and SAS maybe should be changed. Like NCQ is an "additional capability" as they say, nothing about that it should be in SATA/300. Aqualize 22:55, 5 February 2007 (UTC)[reply]

Photo

[edit]

A Hershey's kiss for size comparison?


but they "are one of the most popular candies in the world". Yep, never heard of those before :)


Any chance someone could make a new photo and include a ruler in the picture to show the size/scale? One with metric and imperial would be nice :) It would be a little more useful than using a chocolate as the reference object. —Preceding unsigned comment added by 134.7.248.130 (talk) 05:18, 20 November 2007 (UTC)[reply]

is it ok?

[edit]

List of SAS & SAS RAID controllers - in fact it is a commercial ad with prices, "buy" button etc., not a list. Is it ok in wikipedia.org? —Preceding unsigned comment added by 80.75.94.163 (talk) 12:33, 25 November 2007 (UTC)[reply]

drive identification

[edit]

"SATA devices are uniquely identified by their port number connected to the Host bus adapter while SAS devices are uniquely identified by their World Wide Name (WWN)." so how are SATA drives within a SAS system identified? Plugwash (talk) 18:18, 12 April 2008 (UTC)[reply]

error in SAS layers diagram?

[edit]

In the frame definition layer, I think "SPP transport layer" is a typo; it should be "SSP transport layer". At least I can't find a reasonable definition for SPP, whereas SSP is "Serial SCSI Protocol" :-). --klode (talk) 16:20, 20 May 2008 (UTC)[reply]

I would agree, this seems to be a mistake. (Qualle14 (talk) 22:57, 14 March 2012 (UTC))[reply]

Usage / deployment?

[edit]

I'd like to know what the common usage for SAS drives would be (mission-critical servers, mid-range servers, low-demand servers, etc.) I'd also like to see some numbers as to how many are out there, how many will be out there, etc. Will this technology eventually dominate all server applications? ---Ransom (--208.25.0.2 (talk) 15:56, 3 June 2008 (UTC))[reply]

Please read WP:FORUM, but the short answer is that SAS already has taken over all of the above. -- KelleyCook (talk) 16:22, 3 June 2008 (UTC)[reply]
Could you please cite a reference? It seems to me that, if anything, SATA has taken over all but very high-end servers. Quanstro (talk) 20:27, 19 December 2008 (UTC)[reply]

Is it compatible with SATA 6.0 Gbit/s?

[edit]

Is SAS compatible with SATA 6.0 Gbit/s? Urvabara (talk) 14:56, 27 August 2008 (UTC)[reply]

Yes, they are all (and will be) backwards compatible. Even 6 gig devices when plugged into a 3 gig hba should work... SAS/SATA regardless. It's kind of like PCIexpress (1 & 2) where as either card will work in either slot. 208.69.85.39 (talk) 21:15, 8 December 2008 (UTC) surge[reply]

Slower than Parallel SCSI?

[edit]

In the first paragraph... "At present it is slightly slower than the final parallel SCSI implementation, but in 2009 it will double its present speed to 6 Gbit/s"

I don't think you undestand...it's 3 Gb/sec PER LANE...most SAS HBAs have 2 ports, 4 lanes per port (8 total) -- That's 2400 MB/sec half duplex, 4800 MB/sec full duplex...

Compare that to a Parallel SCSI HBA which has two independent U320 Parallel SCSI Busses (Not full duplex)...640 MB / sec.

I don't see how anyone can say it is slower under any circumstance. 208.69.85.39 (talk) 21:15, 8 December 2008 (UTC)surge[reply]

Number of devices

[edit]

Within the SAS domain section it is stated that a SAS domain can have up to 16,256 devices, while under the comparison to parallel scsi it is stated that sas supports up to 16,384 devices (the latter being 2^14, and thus sounding more correct to me, though I don't know). Are these numbers correct? If so it might be clearer if a note stating this were added. Peter Law (talk) 17:45, 13 January 2009 (UTC)[reply]

port identifiers

[edit]

The article compares sas, scsi and FC port identifiers. It states that the "port identifier" in FC is the WWPN. This seems slightly confusing as the word "Port-ID" is usually referring to the identifier assigned in a fabric or loop to communicate with the port. The WWPN is not used for the communication itself. I think the comparison is fair as long as the "port identifier" is removed. Instead one could say something like HW address.. — Preceding unsigned comment added by Qualle14 (talkcontribs) 22:53, 14 March 2012 (UTC)[reply]

SAS signal pins relative to connectors used in marketplace

[edit]

1. i read that SAS, being differential, and having 4 signals per channel, uses eight signal lines per drive. i read on this wiki page how the sff-8086 (26 pin) connector may be used to transmit four channels. if each drive requires 8 lines, and if four drives are to be supported: 8 * 4 = 32 lines. how is it physically possible to send signals to four drives using only 26 pins? an explanation here would be welcome

2. similarly at the actual drive, the sff-8482 connector is used. the wiki article mentions how on the 'underside' or 'backside' of the connector, additional pins are to be found - without any information as to the pin counts and/or mapping of pin to SAS signal.

3. finally, i realize this page is focused to speak towards the 'standards' side of SAS. however in practice, the market place sometimes offers implementations that might not be fully standards compliant. this is of course then confusing for a consumer. 'sideband' signalling found on many SAS implementations would be one topic that should be included here, so that it is at least discussed. which SAS signal pins are to be used for 'sideband' signal? and how does this need for extra signal relate to the pinouts of the standard connectors sff-8086,8087,8088?

any explanations here would certainly help to remove some of the confusion i now have. thanks. — Preceding unsigned comment added by 70.168.50.55 (talk) 18:20, 27 September 2012 (UTC)[reply]

1.+2. SAS uses one pair of contacts for each direction, so 4 electrical lines per channel. Check www.t10.org/ftp/t10/document.05/05-084r1.pdf for details.
3. Sideband signalling is present to allow for application specific signals without having to use additional cabling (e.g. SGPIO). Sideband pins are present on internal connectors only. Zac67 (talk) 18:51, 27 September 2012 (UTC)[reply]

SAS vs. SSA

[edit]

SAS is reinventing the wheel. All the perks praised in this article were available 20 years ago in IBM's Serial Storage Architecture (SSA):

  • point-to-point links
  • loop topology
  • dual-ported drives
  • simultaneous multi-link transfers
  • multiple initiators (up to 2 in RAID and up to 8 in JBOD configurations)
  • "target mode SSA" communication between initiators

It even had optical extenders which allowed to extend the SSA loop between datacentres at distances up to 10 km.

The only difference being the electronic circuitry at the time was capable of 20 MB/s per link, upgraded in late 90s to 40 MB/s per link. 15 years later the only improvement is frequency increase mere 15-fold. SSA adapters allowed port 1 on adapter A to participate in a loop with port 2 on adapter B with the remaining two ports not sharing a loop. SAS adapters I've seen so far demand port 1 to be joined with port 1 which makes the implementation more error prone.

It seems the WP article on SSA is written by a newbie who knows nothing about it. — Preceding unsigned comment added by 82.152.33.232 (talk) 22:20, 4 January 2014 (UTC)[reply]

SFF-8470 connector pin count

[edit]

I believe the SFF-8470 connector has 34 pins, not 32. I have already changed the article.

Unfortunately, I couldn't find this clearly mentioned in the SFF-8470 standard. It references figure 9, but the references to this figure in the PDF are not links or they do not work. And the figure 9 seems to describe a particular connector variant, not the pin numbering as referenced. But the values (m and n) in the tables correspond as do other verification methods I tried (none of them significantly authoritative though).

The pin counts should be as follows:

  • 16 signal pins,
  • 18 ground pins.

The ground pins are actually 9 dual-sided pairs and could therefore be counted as two per pair, giving 18 pins altogether. Together with the signal pins we get the total count of 34 pins. This is also close to the original number of 32, so the original author might have used the same counting method.

Anyone knowing better is welcome to correct me (and, if possible, leave an explanation comment here). :-) — Preceding unsigned comment added by JITR (talkcontribs) 09:01, 8 October 2015 (UTC)[reply]

Requested move 09 April 2016

[edit]
The following is a closed discussion of a requested move. Please do not modify it. Subsequent comments should be made in a new section on the talk page. Editors desiring to contest the closing decision should consider a move review. No further edits should be made to this section.

The result of the move request was: Moved. I'm not really WP:involved here, as my only action was to query it when it was at WP:RMT, in case someone thought it was contorversial. Hopefully therefore it's fine for me to close this. There is unanimous support to move, so it was indeed uncontroversial.  — Amakuru (talk) 17:18, 19 April 2016 (UTC)[reply]



Serial attached SCSISerial Attached SCSI – It's a proper name, and it's used as such throughout the article. – — Dsimic (talk | contribs) 14:27, 9 April 2016 (UTC)[reply]

@Dsimic, Amakuru, and Dicklyon: This is a contested technical request (permalink). Anthony Appleyard (talk) 22:42, 9 April 2016 (UTC)[reply]
  • Contesting - I think this should go through full RM. I don't have a definite opinion now, but capitalisation/title issues are rarely straightforward, and there are definitely sources that use the current capitalisation.  — Amakuru (talk) 18:13, 9 April 2016 (UTC)[reply]
  • Since it's essentially always capitalized in sources, it's hard to imagine this one drawing any opposition. Just because caps are involved doesn't mean it's potentially controversial. Dicklyon (talk) 21:26, 9 April 2016 (UTC)[reply]
    • Indeed. If some admins are getting worn out sorting controversial from non-controversial upper/lower case conversions, maybe they need to take a break from it, and let others pick that slack up. We seem to have enough admins around for this sort of thing. What we don't want is the entire editorial community bogged down in pointless full-RM discussions that should have been speedied. This would consume far more editorial productivity than a few turned-out-to-be controversial case changes that slip through, especially since WP:RM#Requests to revert undiscussed moves can be used to do those without discussion, and if someone wants to insist on the move, it's up to them to list it for discussion. Treating them as "auto-contested" opens a discussion whether one is needed or not.  — SMcCandlish ¢ ≽ʌⱷ҅ʌ≼  11:32, 17 April 2016 (UTC)[reply]
  • The second and sixth sources in the article itself use lower case for "access", but like I say, I don't really oppose this, I was just querying it. Move it back to uncontroversial if you like, Dicklyon I have no objection to that.  — Amakuru (talk) 22:21, 9 April 2016 (UTC)[reply]
  • Supporting the move: it's a proper name, T10 drafts – probably the most RS – also capitalize the A. --Zac67 (talk) 17:18, 10 April 2016 (UTC)[reply]
  • Supporting the renaming as the one who requested it. — Dsimic (talk | contribs) 21:21, 11 April 2016 (UTC)[reply]
  • Support, since it's the proper name of a specification. If it were not, it would be serial-attached SCSI. We should actually check the sources. If the reliable ones sometimes properly hyphenate the compound modifier, WP should as well, since we always default to proper English, and only make exception for ungrammatical constructions when the incorrect usage is uniform in virtually all RS. If it's variable, we use the proper variant, not the improper one.  — SMcCandlish ¢ ≽ʌⱷ҅ʌ≼  11:27, 17 April 2016 (UTC)[reply]

The above discussion is preserved as an archive of a requested move. Please do not modify it. Subsequent comments should be made in a new section on this talk page or in a move review. No further edits should be made to this section.

missing section MULTILINK (Serial Attached SCSI#MULTILINK) --Tpyvvikky (talk) 13:21, 13 June 2020 (UTC)[reply]

> However, because SCSI commands that can do that are rare, and an SAS link must be dedicated to an individual command at a time, this is generally not an advantage.[10]

The link [10] is to a Tom's hardware piece (authoritative?), and doesn't address the point made that SCSI commands that are duplex are rare.

That statement is from a very non-enterprise POV. As soon as you add expanders, full duplex becomes vital. I'd say we remove it. --Zac67 (talk) 11:14, 22 December 2020 (UTC)[reply]

command reordering

[edit]

In part of another page, there is discussion on SCSI, including command reordering. That is, that you can give a drive a series of commands, such as READs or WRITEs, and the drive will optimize the order, such as based on head or disk rotational position. I didn't see this in the SCSI article, but it seems likely to apply to newer disks, so I though I should ask here. Can disks do this? Gah4 (talk) 22:15, 7 June 2021 (UTC)[reply]

Yes, that feature came with SCSI-2 and all modern SCSI drives support it. See Tagged Command Queueing for details. --Zac67 (talk) 05:58, 8 June 2021 (UTC)[reply]
Thanks, yes, that is the one. It seems that some IBM drives from the 1970's have a feature like this. One is the 2305 fixed head disk, which can have up to 8 requests that it will service in order. It is a fixed-head disk, so no seek time, but there is still rotation. The other is the 3330, where the CPU can read the rotational position and then order the requests. Gah4 (talk) 09:03, 8 June 2021 (UTC)[reply]
Interesting – but not really related to (Serial Attached) SCSI. Possibly interesting for TCQ history though. -Zac67 (talk) 10:40, 8 June 2021 (UTC)[reply]
Oh, yes. I thought, though, you might want to know where the question came from. Gah4 (talk) 21:50, 8 June 2021 (UTC)[reply]

What is a "initiator-target connection"

[edit]

In the "Comparison with parallel SCSI" section it says

"SAS allows a higher transfer speed (3, 6 or 12 Gbit/s) than most parallel SCSI standards. SAS achieves these speeds on each initiator-target connection, hence getting higher throughput, whereas parallel SCSI shares the speed across the entire multidrop bus."

I'm guessing that the "initiator" is the disk controller and the target is a set of drives with it's own SAS connector to the disk controller. But generic words used to reference something specific in a standard, shouldn't be used without reference. BenjaminGSlade (talk) 13:03, 4 October 2022 (UTC)[reply]

Please see the #Introduction section for initiator and target. --Zac67 (talk) 14:31, 4 October 2022 (UTC)[reply]

What is SFF-8480?

[edit]

The description for SFF-8680 says "same pinout with SFF-8480" - but SFF-8480 is not in the table of connectors and there is no further elaboration. This seems like a mistake, as the real SFF-8480 is titled "HSSDB9 (High Speed Serial DB9) Connections" and has nothing to do with SAS. Perhaps it's supposed to read "same pinout with SFF-8482"? That's the only thing that would make sense in context. Mattventura (talk) 01:47, 26 June 2023 (UTC)[reply]

That's really a typo and must be SFF-8482 as you've suspected – corrected. --Zac67 (talk) 04:57, 26 June 2023 (UTC)[reply]

The reverse, connecting SAS drives to SATA backplanes, is not possible.

[edit]

The article says: The reverse, connecting SAS drives to SATA backplanes, is not possible, which I believe is correct. However, we don't know which ones are SATA backplanes. As well as I know, most commodity PCs have HBAs that support SATA and SAS, and will commonly configure to SAS, if the OS supports it. This is easy to see with Linux, as they get names like /dev/sda. On the other hand, I have a Synology NAS device, which as well as I know is SATA but not SAS. But what, exactly, is a SATA backplane? The Synology box I have supports four hotswap drives that plug directly into a PC board. Gah4 (talk) 18:32, 23 July 2024 (UTC)[reply]

most commodity PCs have HBAs that support SATA and SAS' – definitely not. SAS HBAs may be common in servers or even workstations, but they're a long way from being ubiquitous. Modern SoCs integrate SATA (and PCIe for that matter) but SAS always requires additional hardware. A SATA backplane accepts only SATA connectors – there's a land between power and data contacts, so it won't accept a SAS drive. --Zac67 (talk) 18:47, 23 July 2024 (UTC)[reply]
Maybe I said it wrong. Many PCs that I know, which have a SATA drive in them, and seem to be SATA, actually run SAS. I have a nice little Dell running Linux, which the disk mounted as /dev/sda. I believe that means it is running SAS, though I suppose you can't be sure that it isn't emulated SAS. I do have some actual SAS drives, which I could try connecting up to something. I should look up the SoC chip set. Gah4 (talk) 01:23, 24 July 2024 (UTC)[reply]
It seems that I have an OptiPlex 7050, with the Intel Q270 chip set. If I understand the documentation, it can run in SAS mode or SATA mode. As well as I know, it is reasonably typical for an Intel PC. I suspect backplane means one that supports more disks, such as the Synology devices that I have. I haven't checked for the connector land, though. Gah4 (talk) 07:08, 24 July 2024 (UTC)[reply]
The Q270 has no SAS support, see [1]. If there's a SAS backplane, a SAS drive won't work without an extra SAS HBA card or onboard chip. I think you're confused by the IRST RAID mode reporting a SCSI drive (as do most other RAID solutions). That's only logical though, physically it's all SATA. A storage backplane is the back wall of a drive cage, where you plug the drives without additional cabling. --Zac67 (talk) 07:54, 24 July 2024 (UTC)[reply]
OK, I will look at that one. I had an Intel page, but not a data sheet, that said SAS. The small-format Dell PC does hold the (one) drive without cables. Is it only a backplane with more than one drive slot? Can ones with a backplane do data transfer to more than one drive at a time? Gah4 (talk) 18:48, 25 July 2024 (UTC)[reply]
Maybe I am, again, asking the wrong question. I have known for a long time that ATAPI sends SCSI commands through the IDE or SATA bus. It seems that USB also uses the SCSI command set. So, do SATA disks always, sometimes, or never, use SCSI commands? If they do use SCSI commands, it makes sense for the OS to treat them as SCSI. Gah4 (talk) 19:46, 25 July 2024 (UTC)[reply]
A backplane with a single slot? I guess so, as long as the drive is plugged into the backplane, rather than attaching a cable to a mounted drive. Yes, of course multiple drives on the backplane can be access 'at the same time' – fully true with SAS (without expander) and SATA (without switch), only on a slightly larger time scale for parallel SCSI shared bus or through an expander/switch.
ATAPI encapsulates SCSI commands over a PATA/SATA interface. Unrelated to SAS and only commonly used for optical disks, never for HDDs/SSDs (which don't understand SCSI in any way). --Zac67 (talk) 04:35, 26 July 2024 (UTC)[reply]
For many (many!) years, back to mainframe days, disks could seek in parallel, but often only one actual doing data transfer. For ones like IBM S/360, each channel could only do one transfer at a time. Gah4 (talk) 21:07, 26 July 2024 (UTC)[reply]
It seems that SATA disks work in either IDE mode or AHCI mode. Also, both Windows and Linux use the SCSI device driver, and not the IDE driver, for SATA/AHCI devices. It isn't easy to follow the descriptions of the AHCI command set. It might be that they use some SCSI commands, but not as many as I thought originally. But close enough to use the same device driver. Gah4 (talk) 21:07, 26 July 2024 (UTC)[reply]