Meausuring QD in win7

[Anvil]

can we see some charts with write performance?

I'm working on it but I need some more data from you.

I need your 4K random write scores for QD 16 and 32.

Preliminary results for 4K random write

iops__write_qd_single_vs_raid0_9211_x2_qd32.png

update:
- added 4 x X25-M raid-0 WBC disabled
- added 1 x X25-E
- updated Tiltevros QD16-32

[Tiltevros]

the first pic is at 16queue depth @ 4k file random
the second pic is at 32queue depth @ 4k file random

[Tiltevros]

santa came today to my office and quess what :P 1 x25-m was inside a box i think now the 9260 with 8 ports will be my next order come on boss give me my paycheck!!!!!

[Anvil]

Tiltevros,

I've updated the chart.

I can't see why you should be getting the 9260 , your 4K random results are great.

I'd say, wait for a review of the Areca 18xx series if you consider buying another raid card.
Don't get me wrong, the 9260 is a great allrounder, it can't compete with the 9211 on the random small block sizes though.
An improved firmware might change things but at the moment the 9211 is

[Tiltevros]

LSI controllers are optimised for Intels SSD.....

[Anvil]

I wouldn't know about that, you may be right.

My next SSD will probably be the Micron/Crucial C300, let's see how it works on the LSI cards. (hopefully within a month or so)

[One_Hertz]

Well, looking at all the data available online, my current opinion is that Flash based SSDs can not possibly load games any faster (anything measurable) than what we are doing right now because their access time (and thus low queue performance) is already limited by the Flash chips themselves, at around 0.08ms. Next gen SSDs are certainly not looking to be much faster at loading games.

Dozens of places seem to basically confirm this, the only artifact being Napalm's system, which loads much faster (30-40% faster) than the rest of us for no apparent reason.

[Levish]

Well, looking at all the data available online, my current opinion is that Flash based SSDs can not possibly load games any faster (anything measurable) than what we are doing right now because their access time (and thus low queue performance) is already limited by the Flash chips themselves, at around 0.08ms. Next gen SSDs are certainly not looking to be much faster at loading games.

Dozens of places seem to basically confirm this, the only artifact being Napalm's system, which loads much faster (30-40% faster) than the rest of us for no apparent reason.

I think you are looking in the wrong places, or places that don't know what they're talking about with respect to storage performance.

SSD's can definately load things faster, particularly SATAII vs upcoming SATA6Gbps drives. Load times are not simply a matter of IOPs, but also of bandwidth and performance across different file sizes.

Then you get into drives that bypass SATA altogether like the Fuzion IO.

There really is no limit until you start bumping into PCI-e limitations, .1ms is quite low

Napalm's system is pretty tweaked, and optimized with a controller that excels at SSD (or is it simply SATA drives vs ones tuned for both SAS and SATA) and possibly MFT.

[One_Hertz]

A few points to consider:

I have never seen a game use more than 200MB/s in raw read bandwidth using a few different tracers, even though my setup can do 800+

FusionIO shows similar performance to one X25-M in respect to loading games, while being loads faster in benchmarks (lowfat)

The new SF-1500 controllers, which are faster than X25-M in benchmarks, show identical performance to X25-M in game loading (anandtech)

X25-E, which is a lot faster than X25-M in a lot of benchmarks, shows identical loadings times to X25-M (anand).

Raid-0 does not improve game loading times (independent testing from loads of people on this forum)

Acard is noticeably faster at loading games than all current SSDs due to its lower access times. (any review)

Different RAID controllers (between the good ones anyway) hardly have any impact on game loading times (SteveRO)

New gen SSDs can not improve low queue performance until the Flash memory itself gets faster in terms of access times. They can keep making controllers better, but access times will still be limiting things.

If you look at the big picture, all current top end SSD setups are pretty close to one another in terms of loading times. I don't see this changing until either game coders optimize their games for SSDs or access times are decreased.

[Levish]

A few points to consider:

I have never seen a game use more than 200MB/sec in raw read bandwidth using a few different tracers, even though my setup can do 800+

Your system might do 800MB/sec+ at a 128k+ file size but at the 4k file size (for example) you'd be lucky to see 20MB/sec at 1QD

Such is the way RAID works (I'm assuming you are talking about a RAID0 array with the typical 64k or 128k stripe), with respect to being limited to single drive performance at that file level

Then! That single drive is further limited by flash speed, controller, number of channels used, firmware and who knows what more. Which is why you don't see the x25-m doing 200MB/sec at the 4k size (or any other SSD).

Controller improvements and Flash development/improvements can very much bump that rate up significantly and this is where you will see the improvements in loading times!

Or just use a ram disk if you've got that kind of money @_@
I'm not even sure what the latency is on it but most DDR2/3 systems nowadays should be able to do more than 4000MB/sec although then you could run into CPU limitations (with any cpu).

with respect to acard
http://benchmarkreviews.com/index.ph...1&limitstart=6

shows its 4k random reads going to ~65MB/sec which is 3x the speed of the intel x25-m
So its almost 3x faster than a single x25-m at that file size and alot of games have many files at or near that size.
But they also cap out much sooner and lower at the larger file sizes.

Of course we get back to the original comment, since you are investing that much into ram to begin with, you should have just setup a ram disk and enjoyed much higher performance to begin with.

[Tiltevros]

7 ssds in raid.... i cant see where do u find the latancy in multiple RAID 0

[Anvil]

Where is your new drive?

I'll comment on the result later, I'm busy updating the charts.

[Tiltevros]

its 12 at night here after a difficult day @ work i dont have the energy to set it up today

[Anvil]

I've added the Vertex 120GB (FW1.41) to the 4K random read chart.

The Vertex really is a good performer at QD1 but stops scaling at QD >4.

4krandom_read.png

update
- Added SteveRo's Areca 1231ML Acard's

[Anvil]

Here's the 4K random write chart.
It also includes the Vertex 120GB (FW1.41)

Based on the figures I'd really like to see the iometer results from Computurd on the 9211.

iops__write_qd_single_vs_raid0_9211_x2_qd32.png

[Tiltevros]

put the 8k and 16k results ssd r not only 4k files...

[Anvil]

Tiltevros,

Happy?

iometer Random Read

9211-8i_7x_x25-m_mbps_qd.png

[Computurd]

gah i would love to submit some but i refuse to do write testing

[Tiltevros]

@ Anvil posting my results is not about make me happy... i ask u to post results from all of us to see how the raid works in bigger files than 4k... RAID and SSD is not about 4k performance... databases and windows files even game files are not only 4k and 512B... its nice to see the overoll performance of each array.. arecas ichXr Lsi promise etc...

[Anvil]

gah i would love to submit some but i refuse to do write testing

Understandable
I used to have the same reservations regarding write testing but I just don't care anymore.

@ Anvil posting my results is not about make me happy... i ask u to post results from all of us to see how the raid works in bigger files than 4k... RAID and SSD is not about 4k performance... databases and windows files even game files are not only 4k and 512B... its nice to see the overoll performance of each array.. arecas ichXr Lsi promise etc...

Your results were the only ones I found, we just need more people with differents setups to contribute to this thread.
I'm working on doing tests using the ICH and LSI 9260 but it will take some time before I'm done.

[Ao1]

gah i would love to submit some but i refuse to do write testing

Come on Comp it will be interesting to see your results. This is only a small write benchmark. Your Vertex drives have 50 odds years of life left in them and you are going to dump them soon anyway for the new gen drives.

[Anvil]

@audienceofone

Thats exactly what makes me don't care anymore.
I'll be making the switch to SandForce and/or Micron C300 soon or maybe the next gen Intels if the others aren't living up to my expectations.

btw, I've updated the 4K Random read chart with SteveRo's Acards.

[Tiltevros]

@audienceofone

Thats exactly what makes me don't care anymore.
I'll be making the switch to SandForce and/or Micron C300 soon or maybe the next gen Intels if the others aren't living up to my expectations.

btw, I've updated the 4K Random read chart with SteveRo's Acards.

where???

[Ao1]

@audienceofone

Thats exactly what makes me don't care anymore.
I'll be making the switch to SandForce and/or Micron C300 soon or maybe the next gen Intels if the others aren't living up to my expectations.

btw, I've updated the 4K Random read chart with SteveRo's Acards.

Wow those Acards do really well at low qd, but strangely that does not help speed up game loading times.

It seems that the performance issue at low qds is not the only reason for games to not load any faster.

[Tiltevros]

audienceofone and Anvil can i have ur excels files with ur results plz???