Does RAID0 'bypass' SATA3 limit?

[SeeingSharp]

*I know RAID0 isn't perfectly scaled, but for the sake of my question, let's assume so*

Let's say I have 1 SSD with 500 MB/s read. This is within SATA 3's limitations (6 Gbps/750MB). Now, if I add another SSD (and it scales 100%), is it going to cap at 750 MB/s or will it be 1 GB?

Thank you!

[Mark77]

Unless you're doing something funky like using a port expander or port multiplier off of a single SATA circuit, then, no, you are not limited to the speed of a single SATA interface for your entire RAID-0 array.  Assuming perfect scaling (bad assumption, but we'll let that slide for a moment), you can add the numbers up.

 

On some implementations, you might end up with PCI-E lane limitations to the SATA controller which may be limiting.  Or other limitations on simultaneous throughput through the SATA controllers.  But assuming those aren't a factor, things will be additive.

[Godlygamer23]

It will be capped at 1GB/s maximum for the two SSDs because you're using two SATAIII ports(hopefully).

[SeeingSharp]

Cool, thanks guys.

[Enderman]

sata 3 realistically caps at 550MBps

6Gbps is theoretical max

 

raid 0 uses two separate sata cables to give a combined ~1100MBps

[Captain_WD]

9 hours ago, SeeingSharp said:

~snip~

Hey there SeeingSharp

 

A few words from me:

- SATAIII's limit is actually 600MB/s due to the encoding (same case with SATAII having 300MB/s and SATAI having 150MB/s caps).

 

- This limit is per connection ,meaning per port. If you have two drives in a Striped array on two SATAIII ports your limitation would be about 1.2GB/s so unless the slowest SSD in the array has speeds over 600MB/s, you shouldn't get bottlenecked by the SATA controller. 

 

- As the guys suggested you should be OK in your case.

 

Captain_WD.

[EmeraldFlame]

Just to expand on what others have said. While you can increase speeds with raid 0, you can typically only push so far. Most of the consumer grade SATA controllers I have seen used top out at around 2GBps throughput, the controllers themselves just can't process commands quicker than that. So even if you have 10 sata3 SSDs which should give you a theoretical max of around 5GBps, you typically won't come anywhere near that because the controller just can't handle it. The actual limit varies from controller to controller, but they all seem to be around that 2GBps mark. So 3, maybe 4 in some cases, SSD's in raid 0 will give you pretty much the maximum throughput you will ever get.

More drives then that typically won't see throughput increase, although some tests from PCPer, seem to indicate that it can reduce drive latency/seek times.

[SeeingSharp]

@EmeraldFlame Will having more cores on a processor help increase the processing of the RAID controller onboard? Or is that a separate deal altogether?

[EmeraldFlame]

2 hours ago, SeeingSharp said:

@EmeraldFlame Will having more cores on a processor help increase the processing of the RAID controller onboard? Or is that a separate deal altogether?

Separate thing altogether. I'm specifically talking about the sata controller, which is part of the chipset, not the CPU. It is the hardware that actually sends commands to the drives and they typically can't handle more throughput than 2GBps give or take a little depending on the exact model. You might be able to gain a little more performance with bclk overclocking, but I'm not 100% sure as I don't know if the sata controller actually relies on that, it very well may be asynchronous. Without talking to a board OEM or doing some serious testing, I wouldn't be able to confirm.

[Godlygamer23]

1 hour ago, EmeraldFlame said:

Separate thing altogether. I'm specifically talking about the sata controller, which is part of the chipset, not the CPU. It is the hardware that actually sends commands to the drives and they typically can't handle more throughput than 2GBps give or take a little depending on the exact model. You might be able to gain a little more performance with bclk overclocking, but I'm not 100% sure as I don't know if the sata controller actually relies on that, it very well may be asynchronous. Without talking to a board OEM or doing some serious testing, I wouldn't be able to confirm.

Anything before Skylake on the Intel platform does have the BCLK tied to the storage aspect of the system. 

[Mark77]

5 hours ago, SeeingSharp said:

@EmeraldFlame Will having more cores on a processor help increase the processing of the RAID controller onboard? Or is that a separate deal altogether?

 

On a typical motherboard, the "raid controller" isn't anything but special firmware that coordinates an orderly configuration/handoff of the disks installed on the controller to a software RAID driver in an operating system (or embedded into the int13h firmware for legacy purposes).  Sometimes this kind of RAID is called "fake RAID".  So basically its software RAID.

 

The Linux kernel gives a bit of an idea of the sorts of speeds that an intel "general purpose" CPU processor is capable RAID-wise of (and in this case, the RAID driver is coded to make use of the special instructions originally intended for graphics!):

 

[    0.048136] smpboot: CPU0: Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz (fam: 06, model: 2a, stepping: 07)

[    0.286071] raid6: mmxx1     5621 MB/s
[    0.342726] raid6: mmxx2     6059 MB/s
[    0.399370] raid6: sse1x1    4781 MB/s
[    0.456024] raid6: sse1x2    5799 MB/s
[    0.512677] raid6: sse2x1    9619 MB/s
[    0.569330] raid6: sse2x2   11748 MB/s
[    0.569416] raid6: using algorithm sse2x2 (11748 MB/s)
[    0.569505] raid6: using ssse3x1 recovery algorithm

As you can see, even for a RAID6 (ie: double parity), an i7-2600 CPU can handle processing in excess of the speed of the PCI-E bus.  RAID-0/1/10 basically has no "processing" involved with it (unlike the parity-based RAIDs), so there is no conceivable reason why the CPU would ever be a bottleneck in a RAID-0/1 software or "fake RAID" situation. 

 

Of course, if you had the lowest Sandy Bridge chip possible, and used add-in expansion boards for more SATA ports, and ran RAID-6, then, yes, it is entirely possible that the CPU would become a serious bottleneck for RAID-6 performance.