Forum Vs Naplam - Fasted real world storage solution

[NapalmV5]

meow?

[F.E.A.R.]

*curr*

[XS Janus]

haha

[Peen]

I tried your little 40app test with my Xeon rig, was something like 7-8s and with a 4gb RAMdrive with readyboost was only like 3-4s to load the 40apps

[F.E.A.R.]

@ Napalm

2TB Supertalent Raid-Card with 1,4/1,2GB/s. r/w and Intel IOP348. Fast enough? http://translate.google.com/translat...ie%2F447831%2F

[AceNZ]

why go 8x mlc when 4x slc decimates the world ??

Given the name of this forum, do you really not know the answer to that question?

Frankly, I'm much more interested in building the fastest possible system. If 4x SLC decimates the world, and 8x MLC is even faster, then that's what I want.

get 8x x25m + 4x x25e and see for yourself

If I had the time and money to buy and test one of every possible system, I would. Unfortunately, I have to rely on pre-purchase analysis and feedback from current owners and users instead.

since 4x is enough to saturate the controller 4x is the max i go on the 1231

It's exactly that point which makes me think that a PCIe 2.0 controller has the potential of being faster overall, since they have twice the bus bandwidth available.

the fewer the ssds the better the controller handles/controls.. the lower the cpu load

Really? Do you have any tests to back that up, or is it more of a gut feeling? I suppose zero SSDs would be the lowest CPU load, but it definitely wouldn't be the fastest.

[NapalmV5]

@ Napalm

2TB Supertalent Raid-Card with 1,4/1,2GB/s. r/w and Intel IOP348. Fast enough? http://translate.google.com/translat...ie%2F447831%2F

id be interested in the 192gb slc but iop348 ?

thats sour kraut on ice cream

goddamn sas controllers!

so how they do 1.4/1.2gb/s r/w ? custom iop348 ?

whats the interface between the ssd and the iop ?

[NapalmV5]

Given the name of this forum, do you really not know the answer to that question?

Frankly, I'm much more interested in building the fastest possible system. If 4x SLC decimates the world, and 8x MLC is even faster, then that's what I want.



If I had the time and money to buy and test one of every possible system, I would. Unfortunately, I have to rely on pre-purchase analysis and feedback from current owners and users instead.



It's exactly that point which makes me think that a PCIe 2.0 controller has the potential of being faster overall, since they have twice the bus bandwidth available.



Really? Do you have any tests to back that up, or is it more of a gut feeling? I suppose zero SSDs would be the lowest CPU load, but it definitely wouldn't be the fastest.

look you think you know better than me?.. no problem.. but dont be surprised when and why my 1231/4x x25e decimates your 8x mlc/pcie2 sas

[AceNZ]

look you think you know better than me?.. no problem.. but dont be surprised when and why my 1231/4x x25e decimates your 8x mlc/pcie2 sas

I wasn't trying to assert that the 9260 with 8x MLC would be faster than the 1231 with 4x SLC. I've seen the benchmarks in the other thread. In my environment, small block random I/O rates determine performance much more than sequential I/O. The lower latency on the 1231 makes complete sense.

The original question I was trying to ask is something completely different: how would you build the fastest possible system? First choose the fastest drives and controller. If the 1231 is the best, and if more drives or a PCIe 2.0-based controller won't help, then what's next? Stripe a bunch of them together? Are there configuration tweaks, such as NTFS cluster size, strip/stripe size, disabling cache, or what?

Or maybe switching to a native PCIe-only controller would be better? That should eliminate the SATA latency, which would help a lot with small block random I/O.

[F.E.A.R.]

id be interested in the 192gb slc but iop348 ?

thats sour kraut on ice cream

goddamn sas controllers!

so how they do 1.4/1.2gb/s r/w ? custom iop348 ?

whats the interface between the ssd and the iop ?

I think 1,4GB/s. with cache.

[NapalmV5]

if thats the case.. 1231 does 1.6+gb/s

sas controller just cant beat 1231/1261 no way no how

[NapalmV5]

I wasn't trying to assert that the 9260 with 8x MLC would be faster than the 1231 with 4x SLC. I've seen the benchmarks in the other thread. In my environment, small block random I/O rates determine performance much more than sequential I/O. The lower latency on the 1231 makes complete sense.

The original question I was trying to ask is something completely different: how would you build the fastest possible system? First choose the fastest drives and controller. If the 1231 is the best, and if more drives or a PCIe 2.0-based controller won't help, then what's next? Stripe a bunch of them together? Are there configuration tweaks, such as NTFS cluster size, strip/stripe size, disabling cache, or what?

Or maybe switching to a native PCIe-only controller would be better? That should eliminate the SATA latency, which would help a lot with small block random I/O.

whats next is sata3 pcie2 controllers/sata3 drives.. none out and about yet sata6g+sas6g controller = sata3g performance <- you wouldnt want that..

the other option is sas6g drives + sas6g controller: areca 1800 though 4x/8x sas6g drives cost more than a highend mercedes benz

so at the moment if you want the fastest possible system without the hassle of raid cards/x # of drives theres pcie iodrive/ramsan

ramsan: http://www.xtremesystems.org/forums/...d.php?t=230107

[XS Janus]

I think we can all agree that the card to wait for is a new gen. one but without SAS overhead...
The question is who will be the first and how much cheaper will it be in regard to 1231...

[Nizzen]

if thats the case.. 1231 does 1.6+gb/s

sas controller just cant beat 1231/1261 no way no how

I give you competition with my Areca 1680ix

1680ix rules with SAS drives. I scales all the way up to 1250MB/s with 10x 15k.5 sas drives. And I got over 1gb/s read speed with 5x Vertex.

I tested the latency. You remember. My latency was not so bad

[NapalmV5]

*sigh* i thought we went over this nizzen.. no ?

you dont got 1gb/s read.. 5x 150 = 750mb/s at best

if you wanna be like that.. i do over 1gb/s too.. @ h2benchw and if you prefer everest.. 1.6gb/s

5x vertex? i thought its my 4x slc vs your 6x mlc?

let me make it easier..

nizzen vs napalm
1680 vs 1231
2gb cache vs 2gb cache
5x/6x mlc (vertex) vs 4x slc (jmicron)
942mb/s vs 1070mb/s @ h2benchw
1164mb/s vs 1608mb/s @ everest

5x vs 4x
vs

6x vs 4x
vs


and thats my 100mhz pcie vs your xxxmhz pcie ??

[Nizzen]

I beat you @ average read still in everest :p

[NapalmV5]

u should u got ~6x drives but not 1gb/s.. my avg is higher too due to cache

what pcie ? 125mhz ? higher ?

your h2benchw was @ 5x not 6x and everest @ 6x

nizzen: 5x * 150 ~ 750mb/s hence the 711mb/s @ h2benchw
napalm: 4x * 188 ~ 750mb/s hence the 716mb/s @ h2benchw

so whether you like it or not your 6x * 150 ~ 900mb/s at best

[Nizzen]

LSI + 7xintel

[NapalmV5]

note: block size: 512K

lol show me some real apps on that ^ and then tell me you still beat me

what is up with 194mb/s minimum ??

[Nizzen]

lol show me some real apps on that ^ and then tell me you still beat me

what is up with 194mb/s minimum ??

But do you beat that :p

Did you tested Pcmark Vantage? With youre controller you should beat the WR easy

[NapalmV5]

yeh all i care about is highest bandwidth.. i can get 9260 too if id want to show off high bandwidth numbers.. you aint gonna see 9260 numbers from me

i dont do vista/win7/vantage.. if i wanted to break WRs i wouldve submitted results..

im more into real apps.. since thats what i do 99% of the time

[Nizzen]

yeh all i care about is highest bandwidth.. i can get 9260 too if id want to show off high bandwidth numbers.. you aint gonna see 9260 numbers from me

i dont do vista/win7/vantage.. if i wanted to break WRs i wouldve submitted results..

im more into real apps.. since thats what i do 99% of the time

if i wanted to break WRs i wouldve submitted results..

[AceNZ]

How does the Areca 1231ML do with parity RAID (5 or 6)? Still a top performer, or are the 1680 cards better?

[NapalmV5]

i dont do raid 5/6.. only do raid 0.. but why would raid 5/6 performance be diff than raid 0 performance ??

if id do raid 5/6 id still do it on the 1231 as long as the drives are sata



btw: i just noticed rhys been banned.. what happened ??

[AceNZ]

i dont do raid 5/6.. only do raid 0.. but why would raid 5/6 performance be diff than raid 0 performance ??

Small block writes are slower, since there are two reads and two writes for each block (data and parity both have to be read and written).

Also, the parity calculations themselves take time and can introduce latency.