What happened to Indilinx and....

[capeconsultant]

what other new SSD's are coming out in Q1 besides the Intel G3?

Indilinx was the SSD darling when it came out, hardly to be heard from again. What up? Dave

 

[taltamir]

besides the intel G3 we can expect new controllers from both sandforce and indilinx, and all of them look very very impressive from the rumors thus far.

 

[ochadd]

I think the new smaller memory size, 25nm?, was delayed. If I was a mfg it would be something I'd wait for. Sandforce has their new SF2000 coming but sounds like it's going to have business sized margins built in.

 

[capeconsultant]

Where can I go for any news of Indilinx? They have earned my respect with their NOVA, which I have had a for a long time and has been perfect.

 

[sub.mesa]

Don't forget Micron C400; it's going to be great too. SF2000 and Intel G3; those three are the Q1 2010 third generation SSD controllers.

I'm not too impressed with Indilinx so far. They seem to be a budget product that focuses on sequential I/O primarily, with quite poor random IOps. So it's like Samsung a controller for cheaper SSDs running Windows or other light tasks. Sandforce has been making good progress with their new 6Gbps controller it seems; would love to see them benchmarked properly and compared.

 

[capeconsultant]

Well, simply by passage of time Indilinx is now the oldest non updated controller out there. Was even before the Intel G2. SO, I am hoping they are coming out with a serious refresh! Gonna Google their web site now and see if there is any info. SandForce and Intel and even Crucial are getting all the glory, but I still think this horse4 race can change quickly.

 

[=Wendy=]

Jetstream from Indilinx is due Q1 2011, SATA 6Gbps and DDR NAND support at upto 500MB/s.

 

[capeconsultant]

Thanks Wendy. I went to the Indilinx web site and saw some good info there but could not post a link as the site is flash or some such garbage. I will be very interested to see their new stuff.

 

[Emulex]

intel will have capacitor for write back cache. the rest of the guys just do dirty writes (no batter/capacitor for write back). So there's alot of mojo when intel unleashes write-back caching.

intel x25-m write-thru caching > most indilinx with dirty non backed write back cache

I'd rather roll safe than dangerous.

 

[capeconsultant]

Emulex that is very interesting. I had no idea that Intel would have the Capacitor. Safe is always good. I have a theory that SSD's will save untold amounts of data over time simply due to less damage in shipping compared to spinners. For other reasons too, such as no moving parts.

I think it is going to be an amazing Q1 for SSD's!

 

[Emulex]

What you also don't know is the defect rates of modern high density drives and SSD's are extremely high out of the box. they can turn that 128gb to a 96 or 64gb quick if at any time in the process the unit is unstable. disable channel or row. kinda like how a core 2 duo becomes a pentium dual core becomes a celeron. chop off the bad part and sell it for less.

all servers have moved from battery backed write cache to flash back write cache using super-capacitor (instead of battery as of nov.1st) - capacitor holds enough charge (doesn't die every 3 years) to copy the raid controller ram to flash - it's tied to the power line so when the power goes null - copy begins (in a simplish way).

so it's not new technology. But man intel is pulling MAD REAL LIFE performance without any write cache? I can only imagine what they could do with ram dedicated to writing back data like the other guys.

Also the other guys' have compression (doubledisk!) and block dedupe to increase speed and wear as well.

If you think about it - encryption and compression are the same - it only makes sense to compress and encrypt - if you do the math there is enough offset to keep it going faster and faster. One maker does this already - i suspect more will.

google CFFSJ or CFFJ2 filesystem wiki - back in the days pre-SD we had to use filesystems to do compress/journal/wear leveling. Think Olympus XD or raw Flash chips - that was done way before consumer/enterprise ssd was being sold.

here's some new reading:
http://en.wikipedia.org/wiki/LogFS

 

[sub.mesa]

SF2000 also gets supercapacitor; and this makes them suitable as ZFS SLOG device. The supercapacitor protects the DRAM chip; flushing it whenever power is interrupted.

Because modern NAND controllers remap logical LBA to physical LBA, they store this in DRAM. So you have the same problem here of filesystems: the metadata ( HPA mapping table ) needs to be in sync with the data (partly in DRAM, mostly NAND). Without supercapacitor, the mapping tables could be outdated when a power failure occurs, and thus the SSD loses alot of allocated writes. Filesystems are designed to handle that, but up to a point. I believe the most disconcerning is out-of-order I/O and ignoring flush buffer commands. Not sure if these two have been addressed in the new batch of SSD controllers. But they do a step forward in data integrity, that's a good thing!

 

[Zap]

Here ya go!