External Areca Raid maxing out at 115mb/s - why?

[peegee]

Hi all: I have 2 Areca Raid systems on my workstation. The external one is a lot slower, and whilst to some extent I expect this - I'm trying to figure out if/where the bottleneck is.

The internal one is an ARC-1222, which has 6x Samsung 2Tb drives in Raid 5, and gives me ~330Mb/s sequential read. All good. (The HD Tach graph shows erratic line up to 600mb/s and burst of 1106mb/s. Average 330mb/s)

The external is an Areca ARC-5040 which has 6x Hitachi 3Tb drives in Raid 6, and hits a cap at 113mb/s sequential read. The graph is a solid straight line at this level throughout.

The internal one is a somewhat higher level card (800Mhz processor vs 400Mhz) with direct connection, but... still the difference is too big. The external box is connected via eSATA cable, that should not be the throttle on this, but it seems the external one can supply the information faster, but is hitting some restriction along the way, or I have something set wrongly.

Any ideas where I should be looking?

 

[adamantinepiggy]

Well RAID6 takes more processing power than a RAID5 setup (2 versus 1 parity disk calculations and writes) and the fact you have the more complicated array on the lower powered setup does not help.

ARC-1222 is an internal card device that is attached directly to a motherboard PCIe slot with extremely high bandwidth capability.
ARC-5040 is "external" device attached through ESATA port (probably some cheap jmicron or marvel controller chip) and is bandwidth/transfer limited by the ESATA port you connect it too.

The tested actual difference in performance between the 2 arrays is not too big at all. Basically with one array, you are letting it go full throttle directly into the PCIe memory lanes with a powerful controller doing less work. With the other, you are performing more work on a lower powered controller and then also attaching it to a single SATA port on the back of your PC.

The bottleneck is how the external arry gets data to your CPU compared to how the internal array gets the data there. One is using a 4 lane super highway while the other is using a dirt trail in terms of bandwidth. There is nothing to "fix" or compare because the two are not similar setups (e.g. 2 internal PCIe controller cards vs one internal and one external as you have setup).

You don't state your setup, but the only recomendation I could provide to improve performance for the external array is to try attaching it to an Intel or AMD chipset SATA port (if it's not already) instead of using the ESATA port. Maybe even try a USB3 port instead (however I have not tried any USB3 stuff so I can't tell you if this is good or bad). Why? Because on most motherboards, the ESATA ports are generally using a seperate cheap chip controller that does not perform as well as the native SATA ports .. And FYI, it still wont be anywhere near as fast as your ARC-1222 internal RAID controller.

 

[peegee]

well, thanks, but I kind of knew that.

the point is an eSATA has a very high bandwidth - much more the 130mb/s so it shouldn't be that.
and I've had Raid 6 on similar controllers, and got more than 130mb/s. so its not the Raid 6 extra computations.
and the drives are better/faster in the 'slower' array, so its not those either.

something is capping it at exactly this point. the sequential read flat line - a total lack of peaks and troughs means it is hitting a buffer of some sort - a controller or whatever. would be nice to know what it is that is constricting this.

 

[adamantinepiggy]

well, thanks, but I kind of knew that.

the point is an eSATA has a very high bandwidth - much more the 130mb/s so it shouldn't be that.
and I've had Raid 6 on similar controllers, and got more than 130mb/s. so its not the Raid 6 extra computations.
and the drives are better/faster in the 'slower' array, so its not those either.

something is capping it at exactly this point. the sequential read flat line - a total lack of peaks and troughs means it is hitting a buffer of some sort - a controller or whatever. would be nice to know what it is that is constricting this.

Unfortunately, the bottle neck restriction, might simply be the outgoing ESATA port on the external box itself. Between the two sides of an SATA cable connection, one or the other side is only as fast in data transfer as the other side. That's why I suggested maybe try different SATA and/or USB3 port. If there is a restriction on the outgoing SATA port on the ARC-504, perhaps attaching it through another means might help diagnose where the slowdown is. If you try different SATA ports on the PC and they all register exactly the same bandwidth transfers, then it's the outgoing SATA port (or other internal limitation) on the 5040 device that is more likely the bottleneck and you probably wont be able to do anything about it if it's a built-in limitation). Since you can't probably do anything about how this port is setup, then you only other option is to try the other types of connections. USB3 or firewire to see if it bypasses whatever is causing possible hardware restrictions as these two also have higher theoretical bandwidth than what you are currently getting. If all the attachment ports max at about the same, then perhaps it's simply the limitations of the way the external device was designed and no tinkering on the PC side will help matters and you can focus on the external device's RAID setup..

 

[Emulex]

alot of folks don't realize their esata is a usb bridged port (either side).

on my qnap it is, performance is horrific