Talk:Port multiplier

compatability
do port multipliers work with most/all host controllers or only with high end ones?
 * The sata and ahci specs state that port multipliers are an optional feature. Multipliers will not work with all host controllers. Only a few budged controllers/chipsets support port multipliers and then it is usually the slow CB-based and not FIS-based support. It also seems (though I can't find confirmation) that SAS host controllers support only SAS Expanders. They do not support SATA Port Multipliers - only SATA drives. —Preceding unsigned comment added by 193.166.42.19 (talk) 07:36, 11 August 2008 (UTC)

A list or table of support per chipset would be very velcome. I remember reading recently that some new SATA chip (Intel I think) supports FIS. --Xerces8 (talk) 13:44, 22 September 2009 (UTC)

I did receive confirmation that Adaptec raid controllers does not support SATA port multipliers (email from their helpdesk). I am also very curious as to what raid controllers actually support PMP technology and as per previous comment will welcome such a list. From researching the topic to some extend, it appears as if the Highpoint RocketRAID 27xx cards does support PMP indeed. — Preceding unsigned comment added by Skouperd (talk • contribs) 14:54, 28 January 2011 (UTC)

Number of drives
The number of drives that can be realisticly supported is unreferenced, is going to keep changing and IMHO probably should be removed. But if we are going to keep it, we should at least make it realistic. The fastest drives nowadays can achieve up to 120mbps. Many drives in the past 2 years or so have achieved 80mbps+. 4x75mbps drives are already capable of saturating a SATA300 link so IMHO 3 drives makes more sense Nil Einne (talk) 12:59, 12 May 2008 (UTC)

how many device connect to PMP
Send READ PORT MULTIPLIER Command get GSCR[2] value, know how many device on PM. GSCR[2] : 3-0 Number of exposed device fan-out ports —Preceding unsigned comment added by JohnZhang en (talk • contribs) 08:26, 20 August 2008 (UTC)

You mean in RAID0 and perfect straight seq IO and all Data is always at the peripheries for peak throughput. Ok.

Most people interested in Port Multipliers will be planning on RAID-5. At 90MBPS peak each, I am guessing that 4 drives should be ok on a 300MBPS link. Theoretical Max Seq Read for 4xRaid5 is about 360MBPS. That's assuming perfect conditions. In the Real world, it would be under 300 I guess, UNLESS you have an expensive controller and your pattern is ONLY Seq Read. (Backup? Warehousing?)

The more realistic estimation in the majority of real world Raid applications is likely to be on IOPS. The best drives with 150avg IOPS will be 5x150 = 600IOPS for a 5-drive array link. That translates to about 75-100MBPS total, assuming 15K rpm drives with a supersized 256KB stripe. (250 Rotation/s x 256kb) So random I/O will easily be much lower and even 5 drives may be ok? Note that the are physical upper limit calculations, real world applications have many other additional factors. All Peak parameters should probably be halved for average calculations.

Here is an example I found of how much difference controllers can make. A 7x Barracuda ES2 (150 Peak IOPS, 105MBPS Peak) array, FULLY tuned all params, XFS.

* Areca 1280 : 250MB/s write, 350MB/s read, 21000 file created/s * Adaptec ASR52445 : 240MB/s write, 350 MB/s read, 18000 file created/s * 3Ware 9550 : 310MB/s write, 410MB/s read, 6500 file created/s * 3Ware 9650 : 440MB/s write, 410MB/s read, 4500 file created/s

144.230.63.56 (talk) 19:38, 27 October 2009 (UTC)AmitNaidu

This original research really has no place on wikipedia. I suggest the text mentioning 3 drives be removed, as it is completely irrelevant and useless. Even two relatively inexpensive Intel X25-M SSDs in sequential read will go over 300MB/s thus getting throttled by such tech. —Preceding unsigned comment added by 193.153.240.116 (talk) 23:28, 6 January 2011 (UTC)

Benefits/Lower Cost
I think the line stating cost as a benefit should be removed or at least qualified heavily, because it just isn't true. The page referenced never says it is cheaper to use port multipliers, it only says that it is "cost effective". It mentions that you save the cost of the extra controllers by using the PM, but that is only half the story, because you incur the added cost of the PM, and these cost far far more than the controllers. This is because PMs are extremely niche items whereas controllers are commodity items. It is far cheaper to buy a motherboard with more sata connectors on it than to buy a port multiplier, or you can just build an entirely new computer and network it to the first. PMs are really only good for people who do not have the confidence to perform internal upgrades on their computer (or can't because it's a laptop etc), or those who require extremely large numbers of drives connected to one machine. Without buying expensive parts you can connect 16-20 sata drives to a single machine for a fraction of the cost it would take to do it with port multipliers. Unfortunately I don't have any 3rd party references for this, other than ebay.com, but there are also not any references stating that PMs are cheaper. "Cost effective" is marketing hype, and could include saving the cost of paying IT specialists to come to you location and perform the upgrades etc. In the real world PMs are more expensive. --Jackd88 (talk) 22:51, 1 May 2009 (UTC)

Port Multipliers themselves are not very expensive. An eSATA "hub" that sits on a desk just like a USB or Ethernet hub can be had for under $100. The items that are on eBay are full blown multi-bay enclosures and are priced as such. 71.98.253.194 (talk) 17:35, 20 April 2011 (UTC)

Failures
An article from ZDNet reports on the result of failure tests when a port multiplier is used. In short, a single drive failure can prevent accessing any drive using the multiplier. — Preceding unsigned comment added by Jhansonxi (talk • contribs) 04:19, 29 January 2017 (UTC)


 * Sadly, the research paper doesn't detail whether a simple multiplier or a FIS PM was used – while the former can be expected to show this kind of problem, the latter shouldn't but could as well. --Zac67 (talk) 16:33, 29 January 2017 (UTC)