Performance Difference: SAS3 and SAS2 HDDs - What am I missing?

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

XeonSam

Active Member
Aug 23, 2018
159
77
28
So everyone understands the SAS3 interface doubles the bandwidth of the previous generation.
And that a single hard drive regardless of interface will have a sustained sequential considerably lower than what SAS3/SAS2 maxes out on.

I've been telling everyone that the interface doesn't matter for HDD's unless you're dealing with dozens of drives striped for performance. Sustained performance will be similar regardless of raid controller, the drives cache, etc. The speed at which the platters spin will ultimately determine the speed of drive(s).

Then I received a benchmark test with 4 X HGST 6TB SAS3 7200 RPM HDD's which showed sequential read and writes at 6-7GB/s! It was from an HP DL360 g9 and the tool used was crystaldisk mark; meaning it was hardware raid with HP's P440ar running on windows. The 4K read and writes was at 400MB! The volume was 17TB so I'm assuming this is a RAID5 configuration.

1651885779793.png

--> According to the person who showed me the benchmark claims that SAS3 HDD's do infact have much better performance than any SAS2 drive. With or without hardware RAID.

Can anyone validate this is true? I am running 8 HGST SAS2 drives at home in my NAS box but am using a SAS3 HBA. I have loads of SAS3 drives but never thought about replacing them until I saw the benchmark. Anywhere you search on the net, no where does it explain how these speeds can be reached and most of the literature explains that spinning platters are irrelevant to the two different protocols (SAS2 vs SAS3) as you'll only get a max sustain of 300MB/s per drive.
 

BlueFox

Legendary Member Spam Hunter Extraordinaire
Oct 26, 2015
2,063
1,482
113
You're only seeing that speed as it's hitting your HBA's RAM cache. Try something bigger than 1GB for benchmarking and you'll see (you have 2GB of RAM, so need to go well beyond that). No hard drive has yet to go over 600MB/s, let alone 300MB/s (though they are getting close), which is actually the max for the original SAS specification.
 
  • Like
Reactions: Bert and Aluminat

aero

Active Member
Apr 27, 2016
346
86
28
54
Also, keep in mind that maximum speeds are only achievable on the outer edge of the platters. Nowhere even close to max speeds at inner platter.
 
  • Like
Reactions: T_Minus

i386

Well-Known Member
Mar 18, 2016
4,221
1,540
113
34
Germany
Can anyone validate this is true?
Yes. This speed can be achieved by using cache on multiple levels: on the controller (this is what BlueFox means) and on the drive itself. Some hdds have now up to 512MByte cache, in a 24x raid 0 that's 12 GByte...
(Down side: the cache on the hdds is not protected which means you can lose the data (which can coruppt the filesystem) in cases of powerloss or dips/brownouts. That's why it's disabled on raid controllers by default)
 

XeonSam

Active Member
Aug 23, 2018
159
77
28
Awesome. So the performance of SAS3 drives is not better than SAS2 drives. We are only seeing these speeds because of the cache on the raid card. This should then mean that the SAS2 drives should perform equally well, correct? I still don't understand why my friend is so sure that SAS3 out performs SAS2.

So for my home setup which is software RAID (I'm using ZFS/TrueNAS) upgrading my SAS2 drives to SAS3 will not show any performance increase... Is this safe to say?

Many thanks in the explanation!
 

XeonSam

Active Member
Aug 23, 2018
159
77
28
And that the SAS3 performance makes no difference from SAS2 HDD's given that the underlying specs are exactly the same?
 

BlueFox

Legendary Member Spam Hunter Extraordinaire
Oct 26, 2015
2,063
1,482
113
Essentially. Cache would be theoretically a bit faster, but not something you'd notice in actual use. Really only makes a difference for SSDs.
 
  • Like
Reactions: aero

james23

Active Member
Nov 18, 2014
441
122
43
52
I’m curious about this too. Mainly in the case of a 24 Bay SAS2 backplane expander, versus a SAS3 backplane expander using enterprise SAS hard drives (not ssds). And using zfs mainly. (I’m aware there’s plenty other factors outside of the SAS interface)

i’m still searching for actual benchmarks and will update this if I find them, but my question more relates to if there is an improvement in latency/IOPS since the Interface is double the speed – in other words similar how if you go from a one gigabit ethernet to a 10 Gb ethernet, the latency (Ping) goes down substantially- ie a Ping clearly isn’t using isn’t hitting the bandwith Max, yet the ping times are significantly lower between 2x10 Gb interfaces vs 2x 1gb interfaces - I’m assuming because there are alot more “time slots” per 1 second on 10g vs 1g vs 100mbit....
 
Last edited: