eSata Port Multipler Speed Constraint?
in [Home Built PC]
|
From: Smarty on 29 Jan 2010 02:32 I am considering either a port multiplier or an additional eSata controller card in order to connect 4 external eSata drives. Is it neccesarily true that a port multiplier will sacrifice speed when using multiple drives simultaneously whereas separate eSata controller cards ensures faster transfer rates? It would seem that a port multiplier must share the bandwidth of an eSata port for whatever number of drives are connected whereas separate controllers should work faster, up to whatever the bus bandwidth limits may be for the controller (PCI, PCI Express, etc.). As a related question, it would seem that if a port multiplier does indeed limit speed compared to an additional controller card, then why are the port multiplier more expensive than the controller cards, in most cases several times the cost? Why would anybody want to not only get less performance but also spend a lot more for it? Maybe I am missing something here? Thanks for any advice and opinions.
From: Paul on 29 Jan 2010 16:06 Smarty wrote: > I am considering either a port multiplier or an additional eSata > controller card in order to connect 4 external eSata drives. Is it > neccesarily true that a port multiplier will sacrifice speed when using > multiple drives simultaneously whereas separate eSata controller cards > ensures faster transfer rates? > > It would seem that a port multiplier must share the bandwidth of an > eSata port for whatever number of drives are connected whereas separate > controllers should work faster, up to whatever the bus bandwidth limits > may be for the controller (PCI, PCI Express, etc.). > > As a related question, it would seem that if a port multiplier does > indeed limit speed compared to an additional controller card, then why > are the port multiplier more expensive than the controller cards, in > most cases several times the cost? Why would anybody want to not only > get less performance but also spend a lot more for it? > > Maybe I am missing something here? > > Thanks for any advice and opinions. One of the unknowns with PM boxes, is which motherboard SATA ports do they work with ? Until someone posts a comprehensive table of tested hardware, I wouldn't buy one. The only thing listed as compatible, is the SIL3132. See the photo here, depicting a connection to a PM. http://www.siliconimage.com/products/product.aspx?pid=32 To see one of the reasons a SIL3132 is a candidate, see PDF page 22 here. The software interface, has room to list ten disks. The SIL3132 is a two port device. So this shows they're ready to connect two five port PM boxes to the SIL3132 and list the ten disks connected to those boxes. How many other motherboard ports, would have such a capability in software ? (That is called, being "PM aware".) I'd expect to see that in a RAID add-in card, maybe, but motherboard software isn't usually set up for such exotic applications. http://www.siliconimage.com/docs/SATARAID5-UserGuide_v1.30.pdf To date, I still haven't run into anyone who has tested one of those five port boxes. This is the one I'm thinking of. http://www.sataport.com/ And yes, I think your bandwidth sharing theory, is how it works. There is some deal about "frame switching", to communicate which disk they want to talk to. The SIL3726 is a 1 to 5 port PM that needs the host to be PM-aware. http://www.siliconimage.com/docs/SiI-DS-0121-C1.pdf The SIL4726 is a 1 to 5 port PM, with added functions to make it more useful with non-PM aware devices. This might be more amenable to external disk enclosure designs, as well as building plain PM boxes. Note though, that inside chips like this, is a micro doing protocol conversion (for the RAID functions). When you're using the RAID function, and the host port looks like a single disk, the micro that performs the "SteelVine Storage Processor" function, has limited performance. On one of their other chips, they actually stated the performance level as 110MB/sec. I don't know if this one improves on that design or not. (It does. Near the bottom of the page, it says they can do 220MB/sec in RAID0 mode.) http://www.siliconimage.com/docs/SI_4726_SteelVine%20Storage%20Processor.pdf I believe when the SIL4726 is functioning as a PM, the function is purely mechanical, and the processor would not be in the loop. The hardware paths used should look like the block diagram in the SIL3726 datasheet. Paul
From: Smarty on 29 Jan 2010 17:53 Paul wrote: > Smarty wrote: > > I am considering either a port multiplier or an additional eSata > > controller card in order to connect 4 external eSata drives. Is it > > neccesarily true that a port multiplier will sacrifice speed when > > using multiple drives simultaneously whereas separate eSata > > controller cards ensures faster transfer rates? > > > > It would seem that a port multiplier must share the bandwidth of an > > eSata port for whatever number of drives are connected whereas > > separate controllers should work faster, up to whatever the bus > > bandwidth limits may be for the controller (PCI, PCI Express, etc.). > > > > As a related question, it would seem that if a port multiplier does > > indeed limit speed compared to an additional controller card, then > > why are the port multiplier more expensive than the controller > > cards, in most cases several times the cost? Why would anybody want > > to not only get less performance but also spend a lot more for it? > > > > Maybe I am missing something here? > > > > Thanks for any advice and opinions. > > One of the unknowns with PM boxes, is which > motherboard SATA ports do they work with ? Until > someone posts a comprehensive table of tested > hardware, I wouldn't buy one. The only thing > listed as compatible, is the SIL3132. See the photo here, > depicting a connection to a PM. > > http://www.siliconimage.com/products/product.aspx?pid=32 > > To see one of the reasons a SIL3132 is a candidate, see > PDF page 22 here. The software interface, has room to list > ten disks. The SIL3132 is a two port device. So this shows > they're ready to connect two five port PM boxes to the SIL3132 > and list the ten disks connected to those boxes. How many other > motherboard ports, would have such a capability in software ? > (That is called, being "PM aware".) I'd expect to see that > in a RAID add-in card, maybe, but motherboard software isn't > usually set up for such exotic applications. > > http://www.siliconimage.com/docs/SATARAID5-UserGuide_v1.30.pdf > > To date, I still haven't run into anyone who has tested > one of those five port boxes. This is the one I'm thinking of. > > http://www.sataport.com/ > > And yes, I think your bandwidth sharing theory, is how it works. > There is some deal about "frame switching", to communicate which > disk they want to talk to. > > The SIL3726 is a 1 to 5 port PM that needs the host to be PM-aware. > > http://www.siliconimage.com/docs/SiI-DS-0121-C1.pdf > > The SIL4726 is a 1 to 5 port PM, with added functions to make it > more useful with non-PM aware devices. This might be more amenable > to external disk enclosure designs, as well as building plain > PM boxes. Note though, that inside chips like this, is a micro > doing protocol conversion (for the RAID functions). When you're > using the RAID function, and the host port looks like a single disk, > the micro that performs the "SteelVine Storage Processor" function, > has limited performance. On one of their other chips, they actually > stated the performance level as 110MB/sec. I don't know if this one > improves on that design or not. (It does. Near the bottom of > the page, it says they can do 220MB/sec in RAID0 mode.) > > http://www.siliconimage.com/docs/SI_4726_SteelVine%20Storage%20Process > or.pdf > > I believe when the SIL4726 is functioning as a PM, the function is > purely mechanical, and the processor would not be in the loop. The > hardware paths used should look like the block diagram in the > SIL3726 datasheet. > > Paul -- Thanks for your helpful reply, Paul. Given all that you have provided / said and my own research, I am quite certain that my best approach would be to merely add another PCI card with additional controllers. The port multiplier just does NOT look like a good idea, all considered. Thanks once again!
|
Pages: 1 Prev: Your recommendations please Next: Win7 compatibility assistant |