SATA (1.5 gbps) vs. SATA II (3.0 gbps) v.s SATA III (6gbps)

[mattlach]

Hey all,

I've done a lot of googling, but haven't found anything specifically addressing how the SATA interface speed impacts overall array speed.

My initial thought was that it ought to be marginal, as these speeds are "per channel" (right?), so even 1.5gbps sata should be good enough for up to 192MB/s for a single drive, but I'd appreciate some thoughts from those of you with more experience on this topic.

I am using an IBM M1015 flashed with LSI IT firmware, but I just found out that the SAS Expander built into the backplane of the used server I bought, MAY downgrade speeds when SATA drives are attached.

I've been running my drives connected directly to the SAS controller in a single 8 drive RAIDz2 volume previously and have been happy with the performance, but I am planning on fully utilizing the 12 bay expander/backplane in my new (to me) server and doing a pool of two 6 drive RAIDz2 volumes. Just not sure what to expect if the bandwidth of the controller actually drops.

Any thoughts on this?

Thanks,
Matt

 

[cyberjock]

Well, considering rotating media can't even saturate SATA 1 speeds, do *you* think it's a problem? ;)

When just 2 disks can provide more bandwidth than Gb LAN, do *you* think it's a problem? ;)

 

[mattlach]

Well, considering rotating media can't even saturate SATA 1 speeds, do *you* think it's a problem? ;)

Well, I'm asking because I don't know how the bandwidth is distributed.

Is it 1.5/3.0/6.0 gbps total per SFF8087 port, or is it 1.5/3.0/6.0 gbps per drive? If it is total for the port, it could wind up being a problem at the lower speeds, if it is per drive, very likely not at any of the speeds.

When just 2 disks can provide more bandwidth than Gb LAN, do *you* think it's a problem? ;)

Don't assume everyone is using a single 1gbit NIC :p

I have 1 10gig for a dedicated storage only network (between servers), and two 1 gig in LACP for my clients, and yes, I actually do benefit from LACP, and have it set up on my managed ProCurve switch. (Well, most of the time they are idle, but I did this to mitigate some problems I had with stuttering when clients randomly had simultaneous requests)

 

[anodos]

Well, I'm asking because I don't know how the bandwidth is distributed.

Is it 1.5/3.0/6.0 gbps total per SFF8087 port, or is it 1.5/3.0/6.0 gbps per drive? If it is total for the port, it could wind up being a problem at the lower speeds, if it is per drive, very likely not at any of the speeds.

SAS achieves 1.5/3.0/6.0 gbp for each initiator-target connection. An M1015 has 8 internal ports with 6gps throughput per port.

 

[mattlach]

SAS achieves 1.5/3.0/6.0 gbp for each initiator-target connection. An M1015 has 8 internal ports with 6gps throughput per port.

Thank you. That is good information.

Now I guess it depends on how intelligently HP's Expander/Backplane uses those ports.

It uses a single SFF8087 connector to the 12 drive backplane, and the expander is built in. I've heard some cheaper ones just split one of the ports, whereas better ones use all four.