SAS 6/iR Bandwidth or Throughput Max

I believe that 3Gb refers to SATAII/SAS speed, ie, 3 gigabits/s per port. So yeah on paper, 8 ports used would mean 24 gigabits/s. However, 24 gigabits/s is 3 gigabytes/s, whilst the card is an x8 PCIe card which has bandwidth of 2 gigabytes/s (each way). So just a read or write operation couldn't top 2 gigabytes/s, however you're gonna have trouble getting anywhere near those figures depending on both the drives you use and the RAID level you use anyway.

---

@techsys wrote:

is EACH drive going to realize 3Gb/s so that a RAID1 would yield 6Gb/s and a RAID5 (three drives) would yield 9Gb/s, etc., so that an 8 lane chip is capable of 24Gb/s?

---

Note that the SAS6/iR doesn't do raid 5 (see here)

<ADMIN NOTE: Broken link has been removed from this post by Dell>

. For raid 5 you'll need a PERC5 or PERC6.

Also, bandwidth does not equal speed. In a raid 1, your performance will be a little less than the speed of a single drive usually as data has to be written to 2 drives, so the controller has to tell 2 drives to perform the write in question.

Then there's the drive itself. These are generally the performance bottleneck; 3Gbit/s equals about 300MB/s, but there's not a single spindle-based drive on the market (some SSDs are coming close I believe, but Dell doesn't offer these at this time) that could fill this bus/bandwidth (Seagate's latest generation of 15k drives boast a sustained 164MB/s (see this (PDF)).

The system dependency probably also depends on how many drive slots are in the system; if you take a Poweredge 1950 with 3.5 inch drives and a SAS6/iR, the controller only uses 2 x 3Gbit/s, as there aren't anymore drive slots.

If you want hyped theoretical numbers, keep searching, if you want benchmarks go here. The last benches go up to the 5E, close enough. IOmeter server 

http://tweakers.net/benchdb/suite/13

I appreciate the feedback so far but I think I've finally found the correct information at http://www.lsi.com/documentation/storage/standard_product_ics/sas/sas1068e_ds.pdf which is an LSI datasheet regarding the specific chip and yields the specifics I sought.

 

The way I read pages 4 and 5, an 8-drive RAID (which is not the same as the term 'aggregate' in the documentation) would be extremely fast at 4GB/s (that's 4 gigabytes, not gigabits).

 

As this is only theorectical, I have still like to see a large/small file transfer benchmark for this chip using the fastest available drives.

 

"would be extremely fast at 4GB/s (that's 4 gigabytes, not gigabits)."

 

Maybe in 10-20 years from now an affordable server motherboard would be able to handle 4 GB of disk traffic, for now you would need a super computer, and some spare millions to afford it.

As is the adapter you have can not do either raid 5 or raid 10, which seriously limits the maximum throughput you can obtain. The results of the benchmarks at the site link I posted are for mostly high end adapters, one generation back, ...speeds have only marginally increase since then, if any. 

 

 

---

@pcmeiners wrote:

"would be extremely fast at 4GB/s (that's 4 gigabytes, not gigabits)."

 

Maybe in 10-20 years from now an affordable server motherboard would be able to handle 4 GB of disk traffic, for now you would need a super computer, and some spare millions to afford it.

As is the adapter you have can not do either raid 5 or raid 10, which seriously limits the maximum throughput you can obtain. The results of the benchmarks at the site link I posted are for mostly high end adapters, one generation back, ...speeds have only marginally increase since then, if any. 

 

 

---

If you read what I said in my other post, namely "...whilst the card is an x8 PCIe card which has bandwidth of 2 gigabytes/s (each way)" you will notice that the 2 gigabytes each way = the 4 gigabytes/s that the .pdf is talking about. The 8 lanes are referring to the 8 lane PCIe interface, not to 8 drive channels.

Techsys...

 

In the link below, on the first page of the pdf.

 

"Assuming sustained data rates of 100 MB/sec, 24 hard disks should saturate the 2,400 MB/sec available to the two SAS wide ports, causing the x8 PCI Express slot to become a bottleneck. But because PCI and SAS bus efficiencies can prevent the buses from reaching their rated speeds, the bottleneck may appear with a greater number of disks: x8 PCI Express buses typically have an efficiency of approximately 80 percent, while SAS typically varies between 50 and 90 percent depending on the workload."

 

"50 and 90 percent depending on the workload"   likely means an average of  55% in the real world.

 

In the area relating to efficiency, this is where you loose your throughput at the disk system; in reality you can't come close to this throughput. The report link below is theoretical, in reality if you even come close to maxing the disk system, nothing else will be running on your server.

On Sunday I just did a quick benchmark of a Perc6/I on a quad core 3.0 server, with a raid 1 and a 4 disk raid 5, all 15k disks; I am very satisfied with the results, as the test indicate it is >3 times faster than a Perc 4 DC SCSI setup.

While benchmarking, the disk system along with its overhead, and the overhead it places on the motherboard comes close to saturating the system; at that point, basically all functions other than the array come to a halt; so in practice if the disk system produces close to 10% of the maximum theoretical throughput on a steady basis,  not affecting the services and the server resident programs to a great degree, I am happy (for now).

With major advances in mobo design, way beyond PCI-E, with no IRQs, perhaps you will see 40 GB of disk throughput going across an afforable motherboard, not greatly affecting the overall performance of a server; unlikely I will not be around then. 

 

http://www.dell.com/downloads/global/power/ps3q07-20070368-Olivarez-OE.pdf

 

The documentation for the chip seems to indicate 8 channels on the card, whereas the 1064 has four channels.