Forum Vs Naplam - Fasted real world storage solution

[NapalmV5]

It is all CPU limited...

You need to move on from the Q9650. I should get a new setup with an i7 860 today. It wasn't that expensive so I decided to give it a shot to see how much faster loading things would be with a faster cpu. Will take me all frigging evening to insulate everything again, get a good OC going, and make a new nlite image. Expecting 4.5+. I'll post how much faster 498 ff loads.

I'll probably hook up my Es to the ICH10R this time instead of the 9211.

i know but the sexcore is one expensive sex

best wishes on your new setup

[NapalmV5]

Ofc it is not. Game loads are low queue reads. All SSDs are the same at that. ACARDs are a lot faster.

Btw in case you haven't seen it, I did some testing of jmicron vs x25-e over a year ago and came to the same conclusion as you:

http://www.xtremesystems.org/forums/...8&postcount=30

benches sure but reality would beg to differ.. they got very hard time against fast ssds

in my case 0 sec diff @ l4d.. havent had the time for other games but will get to that

[SteveRo]

thanks steve.. indeed excellent choice.. i got no problem walking all over you acards @ as ssd

ever since i got the jmicrons 512mb was always faster vs 2gb and later on 4gb @ real apps.. if 2gb/4gb faster thats what i wouldve used.. 2gb/4gb only faster @ benches

yeh they go very well together.. 3x x25e/4gb maxes 1231 out and thats why im @ 3x raid 0

yes, it does appear you can overload the 1231 - 12xacard vs 5x or even 4x - the 5 and 4 R0 seems to be faster in pcmv.

With your new 1231/3x x25e have you tried reducing the stripe size?
I got a good improvement droping from 64K (areca default) to 4k stripe.

[SteveRo]

... make a new nlite image ...

Last time I used nlite it needed a lot of help to work with w7.
What do you throw out what do you keep - I would like to try this again if I had a better idea of what I should keep vs what i can get rid of.

[One_Hertz]

Last time I used nlite it needed a lot of help to work with w7.
What do you throw out what do you keep - I would like to try this again if I had a better idea of what I should keep vs what i can get rid of.

I use XP.

I spent 8 hours trying to get vlite to work with windows 7 so I could get rid of 90% of that bloated crap, but ultimately I could not get it to install. The POS kept saying it could not find my DVD rom as soon as I made any changes. I even tried installing from a USB stick without a DVD ROM plugged into the system and it still gave me that retarded error and aborted the installation. After wasting a full day trying to get it to work I nerd raged, deleted w7, microwaved the w7 install disk, and installed xp.

[NapalmV5]

yes, it does appear you can overload the 1231 - 12xacard vs 5x or even 4x - the 5 and 4 R0 seems to be faster in pcmv.

With your new 1231/3x x25e have you tried reducing the stripe size?
I got a good improvement droping from 64K (areca default) to 4k stripe.

still kept @ 64k but ill load l4d @ 4k stripe and see if its any faster thanks steve

posted a tube loading l4d @ 1080p default settings @ 4x25e/1231/64k stripe

http://www.youtube.com/watch?v=80AUq8CwqeY

no faster than 4x jmicrons

[NapalmV5]

this tube loading l4d @ 3x25e/1231/4k stripe

http://www.youtube.com/watch?v=meDM0RKD7kA

maybe you guys can see diff between the two

[SteveRo]

I use XP.

I spent 8 hours trying to get vlite to work with windows 7 so I could get rid of 90% of that bloated crap, but ultimately I could not get it to install. The POS kept saying it could not find my DVD rom as soon as I made any changes. I even tried installing from a USB stick without a DVD ROM plugged into the system and it still gave me that retarded error and aborted the installation. After wasting a full day trying to get it to work I nerd raged, deleted w7, microwaved the w7 install disk, and installed xp.

If I treated my microwave that poorly it would probably catch my kneaded eraser insulation on fire instead of just sparking all over it!

[SteveRo]

this tube loading l4d @ 3x25e/1231/4k stripe

http://www.youtube.com/watch?v=meDM0RKD7kA

maybe you guys can see diff between the two

wow, I can't tell - their both so fast that you can't really see a difference

[Nizzen]

this tube loading l4d @ 3x25e/1231/4k stripe

http://www.youtube.com/watch?v=meDM0RKD7kA

maybe you guys can see diff between the two

Are you using Firmware 1.48?

Looks like Areca is still good enough

[NapalmV5]

thanks guys.. yeh 1.48

heres one more.. tube loading l4d @ 2x25e/1231/4k stripe

http://www.youtube.com/watch?v=L3MZVDH4P3A

should i go for 1x25e ??

[Anvil]

Of course you should

[NapalmV5]

lol oh man.. alright

tube loading l4d @ 1x25e/1231/4k stripe

still processing : http://www.youtube.com/watch?v=03wUn9pSHVE

well guys i sheat you not thats 4, 3, 2, 1x25e/1231/512mb cache vids loading l4d

[One_Hertz]

the single X25-E is half a second faster than all the RAIDs lol. Surprise surprise, raid doesn't help loading times.

[[XC] Oj101]

How do you explain that?

[FlawleZ]

How do you explain that?

I think it could be explained like this:

You have 4 main types of wait:

Access Time
Read Time
Write Time
Processing Time

RAID has always excelled at providing substantial improvements in Read/Write and minus SSD's, Access Times. With a fast SSD, theres really little to no improvement in Access Time vs a single SSD. So, anything large in amount of access requests are going to be mostly limited by your Processing Time, or CPU.

Now,this would explain something like a batch file of 100 notepads or FF processes, but with L4D, I guess its CPU limited? Remember most of the CPU load is very dependent on a strong IPC ratio. Core count isn't always taken complete advantage of in these types of tests.

[[XC] Oj101]

I would understand that if the times were equal, but a single drive BEATING RAID? Does the card introduce massive latencies or something?

[Computurd]

alot of it is an overloaded IOP as well. the IOP 348 was not designed for anything even close to this type of speed. i believe the 1231 came out before ssd;s were even available.

[AceNZ]

How do you explain that?

In part, due to low QD.

Most people will only see better I/O perf with RAID if they have moderate QD.

[GullLars]

alot of it is an overloaded IOP as well. the IOP 348 was not designed for anything even close to this type of speed. i believe the 1231 came out before ssd;s were even available.

Overloaded IOP, maybe. Benchmarks show it pulling off 50-60K IOPS, 1 x25-E can do around 40K both read and write. Latency overhead on larger RAIDs will probably be in the range 10-100µs pr IO request (dependent on number of devices, stripe size, and QD), wich is minor for 4KB, and negligable for larger blocks.
To do the math, x25-E can do around 5-6K 4KB random IOPS pr channel = 1s/5000 = 0,2ms = 200µs. Worst case would give a rather large RAID 5-10% disadvantage over a single drive.


As for the IOP 348 being made before SSDs, not even close.
http://www.storagesearch.com/chartingtheriseofssds.html

1976 - Dataram sold an SSD called BULK CORE which attached to minicomputers from Modular Computer Systems and emulated hard disks made by DEC and Data General. Each chassis held 8x 256k x 18 RAM modules and had a capacity of 2 megabytes.

>> Acard XD

I'll list a few other entries from the article linked above, just to show some interresting historic events.

In 1978 - Texas Memory Systems introduced a 16 kilobyte RAM-based solid state disk
In the early 1980s - Intel's 1M bit bubble memory created a lot excitement as a new non volatile solid state memory technology.
1985 - Curtis introduced the ROMDISK, the first SSD for the original IBM PC.
In 1987 EMC introduced SSD storage for the mini-computer market
In 1990 - NEC marketed 5.25" SCSI SSDs using internal battery backed RAM.
In 1991 Digital Equipment Corp marketed the EZ5x family of Solid State Disk accelerators.

In 1995 - our SPARC Directory listed 2 SSD products aimed at the Sun server market.

* T8000 - was an 80MB, 10MBps SSD on a single slot SBus card, made by Colorado based CERAM. Units in multiple slots could be chained to appear as a single SSD upto 960M. Performance was 2,000 IOPs.

* SAM-2000 was a rackmount SSD upto 8GB, with 500MBps internal bandwidth- made by Texas Memory Systems. The transfer rate through the SBus adapter was 22MBps. Other bus interfaces included VMEbus and HIPPI.

In 1996 - ATTO Technology maketed the SiliconDisk II. It was a 5.25" form factor SCSI-3 interface RAM SSD with 64MB to 1.6GB capacity. Throughput was 80MB/s, and performance was 22,000 IOPS.

In 1999 - BiTMICRO launched an 18GB 3.5" flash SSD.

In November 1999 - the number of market active SSD manufacturers listed on STORAGEsearch.com had reached 11.

In June 2001 - Adtron shipped the world's highest capacity 3.5" flash SSD. The S35PC had 14 gigabytes capacity and cost $42,000.

[Computurd]

wow they are older than i thought. but i was referring to mainstream usage. i would bet alot of money ssd was not on its compatibility list when released. he needs a faster more robust raid card...1880 maybe?

also of note....
on each of those videos only the first run done is really relevant. on the successive runs the data is already partially cached (program data, textures, etc...) which is why the second third and fourth loads are so much faster than the first. even if different maps are loaded in between. if you close out the program and load those levels from a restart, they would take twice as long, so this is not an entire indicator of the speeds of the drives. only the first level loaded is relevant.

[NapalmV5]

^ blah blah blah.. bravo @ getting more ridiculous/idiotic

third and forth loads brilliant!

i guess the included video clip which is the reason why first load of any level takes longer is relevant.. go cry a river somewhere else

heres another tube this time loading crysis @ 3x25e/1231

http://www.youtube.com/watch?v=rpzAM5LzeBI

your going to like it computurd i promise you

[Computurd]

its fast, dont get me wrong, but here is a challenge to you. make me a video of you loading the game, then after loading the game, directly load the death toll stage. load the stage in under 3 seconds, like it does with successive loads, and you will have proved me wrong. on your L4d level loading video that is how fast it was, after a previous map had loaded.
i have posted L4D speeds that match your load speeds here a few months back on this forum.
also, i have showed how loading other levels, after even a different first stage is loaded, is faster. i restart the application each time i load the levels for a accurate non-cached load speed.
but really who cares, cache is fast and meant to be used, it just means that others can do the same thing however, no matter what speeds their SSD's are, provided they have cache as well.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
those crysis loads are impressive though, nice job. what mod is used on those maps? i noticed that they are not asking for you to "press any key to continue" prompt is not showing up when the load is done like it does on both crysis and crysis warhead. went and found the loader you are using, too, and it still has the :press any key to continue: prompt once the load is done.
also, on your MESA load you can see the load bar go up to 82 percent, then go BACK to 81 percent, then go to 100 percent....strange. i have never seen my game load, then the bar go backwards, then jump to complete. i have never seen the bar go backwards, period.
once the level is done loading, if you could hit the tilde button for us, it will show the level load time that the game itself measures. (and just a note here, mine loads just as fast, the differences are milleseconds, thats why i would like to see the level load time )

-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
back to L4D:
question for you as well, genius. on the first video, with 4 x25e/1231/64k stripe, you loaded five stages, the last stage being no mercy. the first time you loaded no mercy in that string of loads it loaded in 6.70 seconds. the second time you loaded it, even though you loaded three maps inbetween, it loaded in 3.01 seconds. you have already proven what i am saying about consecutive level loads, but with your own video .but if you really want to prove something...answer my challenge, however, i want to see the game loading from the desktop, then straight to the death toll map.

or here would be an even better one, restart L4d each time, so that we can see the load screens for the game, inbetween level loads, then we will compare the times on all the maps you loaded. you might learn something

[antiacid]

start loading some gta4 episodes from liberty city save games, those really are a pain to sit through on reloads

[[XC] Oj101]

Ok, you're ALL right, now less talk and more action please