The $99 dollar SSD has arrived

[glussier]

Thanks for the pictures.

You know, you can upload pictures at http://imageshack.us/

[halfmanhalfamazing]

Pics are up

Finally got them up on a site. I can't hotlink them here, Angelfire sucks like that. I created the most basic of basic pages for display pursposes here:

http://www.angelfire.com/rpg2/tweakit/index.html

[halfmanhalfamazing]

This is the Extreme 3 card:

/sbin/hdparm -i /dev/hdc

/dev/hdc:

Model=SanDisk SDCFX3-4096, FwRev=HDX 4.08, SerialNo=116918E2707D4818
Config={ HardSect NotMFM Removeable DTR>10Mbs nonMagnetic }
RawCHS=7964/16/63, TrkSize=0, SectSize=576, ECCbytes=4
BuffType=DualPort, BuffSize=1kB, MaxMultSect=4, MultSect=4
CurCHS=7964/16/63, CurSects=8027712, LBA=yes, LBAsects=8027712
IORDY=no, tPIO={min:120,w/IORDY:120}, tDMA={min:120,rec:120}
PIO modes: pio0 pio1 pio2 pio3 pio4
DMA modes: mdma0 mdma1 *mdma2
AdvancedPM=no WriteCache=disabled
Drive conforms to: Unspecified: ATA/ATAPI-4

* signifies the current active mode

Hmmmm............ sometimes the answer can be had just by comparing results.

I noticed this line:

Config={ HardSect NotMFM Removeable DTR>10Mbs nonMagnetic }

Also appears in the results for the Extreme IV card. Not being a dev myself, I can only take a stab in the dark on this, but perhaps my system/driver is accurately ID'ing this thing as what it really is........ a removable media/card and bottlenecking me at 10MB/s?

I ran this on the EX3 card:

# hdparm -Tt /dev/hdc

/dev/hdc:
Timing cached reads: 788 MB in 2.00 seconds = 393.73 MB/sec
Timing buffered disk reads: 20 MB in 3.16 seconds = 6.33 MB/sec

I'd bet that's what it is. Wherever that configuration is set at, that's the problem. May not even be driver related; just some text/config file somewhere that's doing it.

[halfmanhalfamazing]

This is the Extreme III card:

/sbin/hdparm -i /dev/hdc

/dev/hdc:

Model=SanDisk SDCFX3-4096, FwRev=HDX 4.08, SerialNo=116918E2707D4818
Config={ HardSect NotMFM Removeable DTR>10Mbs nonMagnetic }
RawCHS=7964/16/63, TrkSize=0, SectSize=576, ECCbytes=4
BuffType=DualPort, BuffSize=1kB, MaxMultSect=4, MultSect=4
CurCHS=7964/16/63, CurSects=8027712, LBA=yes, LBAsects=8027712
IORDY=no, tPIO={min:120,w/IORDY:120}, tDMA={min:120,rec:120}
PIO modes: pio0 pio1 pio2 pio3 pio4
DMA modes: mdma0 mdma1 *mdma2
AdvancedPM=no WriteCache=disabled
Drive conforms to: Unspecified: ATA/ATAPI-4

* signifies the current active mode

[halfmanhalfamazing]

Just as a reference point, here's what I'm expecting from this extreme IV card.

http://www.robgalbraith.com/bins/con...id=7-7896-8475

Most benchmarks put it in the 30MB+/- range, but at the bare minimum it should be 20MB+/-.

[halfmanhalfamazing]

# hdparm -Tt /dev/hdc

/dev/hdc:
Timing cached reads: 788 MB in 2.00 seconds = 393.17 MB/sec
Timing buffered disk reads: 20 MB in 3.10 seconds = 6.45 MB/sec

[halfmanhalfamazing]

Ther's actually UDMA mode 3 activated for that card, which means a maximum transfer speed of 44mB/sec. When you load an application or boot your computer, you are actually loading multiple files, so what is most important is not the transfer speed as such but the latency. With the memory card, it doesn't matter where the files are stored, the latency is pretty much the same, no matter where the files are stored on it, with a hard drive, the controller has to mode the reading/writing heads, which takes a lot of time.

Alright. That does make sense. So while I've got a maximum of 44mb, my actual is going to equate to 6? I'd think that it'd be somewhere higher than that, but sure. I can understand the bandwidth/latency equasion too. Somewhat.

But there's still that issue of having to pass off the IDE=NODMA command at bootloader time.

If I don't do that, the system hangs and I get DMA timeouts. It'll eventually boot, but I could cook dinner in the amount of time that passes by.

[Svartalf]

Agreed. I knew going into all of this that access times were going to be a net benefit, but it wasn't until I saw the transfer rate numbers that I realized just how much of a difference that access times truely mean.

People only see bandwidth- they rarely think in terms of latency. Part of it is due to everyone harping on the bandwidth numbers. A Hummer full of 500Gb SATA HD's has more bandwidth than pretty much any pipe we can concieve of. The problem lies in the latency of the transfers with it. If you're trying to push hundreds of terabytes, it makes complete sense to do it that way because it'll be waaay faster- whether it's across town or across the country because the overall bandwidth exceeds any pipe you can field otherwise. Now, try moving just a CD's worth of data across the country that way. The latency makes it much less worthwhile because all that bandwidth is being left lying on the floor, unused. The same goes for things like disk interfaces, LAN cards, WAN interfaces, etc.

Unfortunately, much of our disk accesses on most machines are more like the CD than the multi-terabyte transfer in the above analogy. That's why "counterintiuitive" things happen like that "low" speed flash disk either over USB or over PATA/SATA seeming to be a LOT faster than the "fast" SATA disk drives we have nowadays.

[glussier]

Yes Sir.

# /sbin/hdparm -i /dev/hdc

/dev/hdc:

Model=SanDisk SDCFX-4096, FwRev=HDX 4.04, SerialNo=012222C1207Q0359
Config={ HardSect NotMFM Removeable DTR>10Mbs nonMagnetic }
RawCHS=7964/16/63, TrkSize=0, SectSize=576, ECCbytes=4
BuffType=DualPort, BuffSize=1kB, MaxMultSect=4, MultSect=4
CurCHS=7964/16/63, CurSects=8027712, LBA=yes, LBAsects=8027712
IORDY=no, tPIO={min:120,w/IORDY:120}, tDMA={min:120,rec:120}
PIO modes: pio0 pio1 pio2 pio3 pio4
DMA modes: mdma0 mdma1 mdma2
UDMA modes: udma0 udma1 udma2 udma3 *udma4 udma3 *udma4
AdvancedPM=no WriteCache=disabled
Drive conforms to: Unspecified: ATA/ATAPI-4

* signifies the current active mode

Ther's actually UDMA mode 3 activated for that card, which means a maximum transfer speed of 44mB/sec. When you load an application or boot your computer, you are actually loading multiple files, so what is most important is not the transfer speed as such but the latency. With the memory card, it doesn't matter where the files are stored, the latency is pretty much the same, no matter where the files are stored on it, with a hard drive, the controller has to mode the reading/writing heads, which takes a lot of time.

[halfmanhalfamazing]

Aaahh... The fun of benchmarking things.

What precisely are you measuring when you set up a metric?

Nothing really, specifically. I'm just replying back to a request from another poster. The only thing it really serves to me personally is to show that something isn't quite correct. I have a feeling that it's driver related, I don't see what else it could be.

Michael's got a fun job of trying to come up with things that actually have meaning when he does his performance comparisons (Which, by the way, keep doing the great work you have been doing in that space... ).

Agreed. I'd like to see a professional Phoronix review on this at some point in the future. I emailed Sandisk about this, not sure what kind of reply I'll get though. But I'm sure that an email from a review site with information requests will carry a little more weight than just some guy who does so. But we'll see. I'll post their reply when it comes through.

In the case of disk performance, data throughput speed IS important, but it's only a piece of the puzzle.

Agreed. I knew going into all of this that access times were going to be a net benefit, but it wasn't until I saw the transfer rate numbers that I realized just how much of a difference that access times truely mean.