SATAII bottlenecking 4k reads/writes?

C

chrcoluk

[H]ard|Gawd
Joined
Jul 7, 2008
Messages
1,260
Ok I got myself a new SSD and its a SATAIII model on the understanding it should still work on SATAII and if I upgrade my rig in future I can take advantage of SATAIII later, Although I have a marvell SATAIII controller I am not using it aware of the issues associated with it, the ssd is connected to my intel SATAII controller which is already in AHCI mode, currently its empty just some benchmarking done on it.

Its a samsung 830 128gig.

I am using msachi drivers.

So here is some test results with my comments.

Crystal diskmark

Seq read 261.8
Seq write 247.1 - Both these seem ok as would expect hitting SATAII max speed.

512k read 205.7
512k write 217.1

4k read 14.58
4k write 46.59 - these seem way below other benchmarks I seen, and not anywhere near the SATAII max bandwidth, so is this something I need to look into or does SATAII slow down everything on a SATAIII drive?

4k qd32 read 171.8
4k qd32 write 82.53 - these also lower than comparisons of same drive.

as ssd same sort of thing, othe rthan sequential dissapointing.

seq read 262.35
seq write 247.06
4k read 15.29
4k write 47.09
4k-64thrd read 179.22
4k-64thrd write 78.82
acc-time read - error cant find drive o_O
acc-time write - 0.064ms

also the 4k-64thrd tests saturated my cpu, these sucked cpu hard, is this normal? I got a 4 core I5 that was otherwise idle during tests. So the cpu bottlenecked the 4k-64thrd benchmark.
 
ok an update incase anyone following this.

I updated my bios, the official latest bios did actually bump the intel option rom from 8.9 to 10.5 to get round the 2TB hdd limits, because that was in official I knew my board probably could handle 11.2 as 10.5 is on Z68 boards. So I grit my teeth and made my own custom rom (first time I done this) this bumped the intel to 11.2 (I didnt use 11.5 as is considered beta?).

When I first booted back up and did more benchmarks I had a 30% drop in 4k reads, every other read test also dropped except threaded 4k reads.

It turns out disabling C-STATE on the cpu has a sigificant impact on ssd performance, I disabled it (as was before bios update) and I got my missing read performance back, however threaded reads dropped slightly as that actually went up with C-STATE on.

My results still seem low, and I still wonder if its due to SATAII (even tho not hitting SATAII throughput limit but maybe it also limits iops?) or due to msachi as this hasnt been tested on intel RST drivers.
 
Last edited:
<<Although I have a marvell SATAIII controller I am not using it aware of the issues associated with it>>

those issues have been fixed at least with sandforce drives from kingston
Plug the ssd into the marvel port but dont use a marvel driver.
try it I think youl like the speed difference
 
more observations.

with c-state off turbo mode doesnt work so peak cpu clock is 2.66ghz instead of 3.2ghz which means I now have lower qd64 results on ass benchmark proving thats cpu limited.

turning c1e off further boosts all the other tests tho (except sequential which already maxed). However 4k is still below what I seen reported by others (seems all others run this drive on sataIII tho).
 
on marvell with 5gbit/sec of pci-e bandwidth activated taken from my graphics card. using msachi.

threaded performance is down to the point cpu is no longer saturating it although it does seem cpu usage is just way down for NCQ and performance is not as down as much as the cpu usage drop, sequential reads are up but not saturating either ssd spec or the 5gbit/sec pci-e limit. seq writes are slower but non threaded 4k writes higher . access time slightly improved. so mixed results

as ssd

seq read 361.62 write 221.82
4k read 22.00 write 64.30
4k-64 read 143.38 write 54.60
acc time read 0.065 write 0.049

interesting msahci works on marvell, does this fix all the reported marvell stability issues? bsod etc.

the issue is with this config is that I have a hdd on the marvell which hates NCQ (marvell was in ide mode) The hdd cant go on the jmicron as that controller seems to run aweful giving all sorts of issues. Also I am unsure of the impact on my graphics card with half of its pci-e bandwidth taken away. decisions I guess.
 
interesting msahci works on marvell, does this fix all the reported marvell stability issues? bsod etc.>>

with the samsung you should be fine, if the disk was an older model than not
 
they not currently used as slowed down my system, and reliability issues with devices not responding, although I may try them out for benchmarking the ssd, I did wanted to see if others had comparisons between msahci and iastor but couldnt find any.
 
Crystal diskmark for the jmicron. will move it back to intel after this and test with 11.2 intel drivers.

Seq read 374.1
Seq write 240.5

512k read 293.6
512k write 188.2

4k read 23.38
4k write 57.36

4k qd32 read 159.3
4k qd32 write 77.31 - again these 2 tests lower cpu usage and not saturated cpu.
 
intel 11.2 is almost the same except bit faster on 4k reads by about 5%, however the system is noticebly sluggier of the hdds(still booting off hdd).

I wont be doing much more testing as I am just wearing the drive down with these tests which use massive 1 gig test files, so soon will get on with my migration and even with these results I guess is way faster than hdd.
 
guys I think overall the marvell port is better than my intel port, seems sataII intel ports are overated for ssd. The question will be if its stable and I think will be on msahci drivers.

I am still not happy with my high queue depth performance but am accepting it for now as it should still rip my hdd apart for speed.
 
You must log in or register to reply here.
Back
Top