256GB SSD raises the bar

"Samsung didn’t mention how much the 256GB SSD would cost." That's why mechanical hard drives, no matter how much slower they are, will still far outsell this device. I imagine street prices for this SSD in the US will be $1,000 - $1,500. Plus, flash still seems to be somewhat less reliable than mechanical. I would rather buy a mechanical drive rated for 1.2 million hours than a flash disk rated for 100,000 operations. Either way, I'd still RAID 1 any data storage volume I use.

i bet it costs a mass-whack they cant compete with regular drives until they bring the price down

I don't think it'll beat Intel's mlc SSD's.

(Yahoo after advancing 40% to $18)shares hit a 52-week low of $9.24 after Ballmer said at a shareholder meeting that a search partnership would be "interesting" but Microsoft had "moved on" from any acquisition deal. Next: it makes sense read & write speed should be 1:1. How can you read if cann't write & conversely, How can you write if read is slow? SSD is much more likely to achieve potential speed spec than HDD, yet NT6 isn't speed demon, so itsgonna take some time. drashek

Initially this will cost as much as an adaptec 8 drive SAS/SATA controller + 8 1TB samsung F1 drives. Assuming you would have such a lovely RAID5 array, you would have similar or better read speeds & about half to three quarters write speeds...not terribly impressive I know. But then again you would have 7TB of space (-1 cause of RAID5) AND you would have fault tolerance. The drive could possibly cost even more than the above mentioned hardware. Bottom line: It's not worth the money, an enthusiast item or application server only. (Could be quite handy with video editing too) Soon these drives will be dirt cheap, give it a couple of years.

If 64GB SSD cost us $230USD, imagine what 4 times of the size would cost, maybe around $1000USD?

With RAID 1, which disk is the good copy? You don't know? With moving parts magnetic storage you can get soft read and write errors that may/may not be recoverable all the time. Typically they're recoverable. But 1 in about 12 terabits (if you believe the vendors...I don't anymore...) will come back as a complete failure. With RAID 1, there's no coming back from a copy failure or you losing the "good" disk. RAID 1 doesn't do what one would think it does. Nor does RAID 5 do it either. SSDs are more expensive and yes they'll still outsell for many applications for some time to come. The big takeaway is that they will still outperform all but the most expensive disks (heh... that gap on price shrank quite a bit...) and they'll survive things that would render the building or space that the machine resides in into scrap/rubble.

"Initially this will cost as much as an adaptec 8 drive SAS/SATA controller + 8 1TB samsung F1 drives. Assuming you would have such a lovely RAID5 array, you would have similar or better read speeds & about half to three quarters write speeds...not terribly impressive I know. But then again you would have 7TB of space (-1 cause of RAID5) AND you would have fault tolerance. " You're forgetting one REALLY important thing... on the vast majority of systems, most filesystem access is for lots and lots of small-medium files. That RAID 5 array will get you great throughput when transferring multi-GB files, but every drive in a RAID set increases the latency. 8 drives * 4.17ms = 33.36ms. SSDs have nanosecond (that's 1/100th of a millsecond) access times. That SSD will kick your big fat RAIDed ass around the room for boot times, gaming, and pretty much any use other than massive databases and DVD storage (full quality DVDs, not ripped. The ripped ones are rarely multi-GB, and many-multi-GB is the only place you'll see a benefit from that RAID).

And I quote: "but every drive in a RAID set increases the latency. 8 drives * 4.17ms = 33.36ms." As far as straight striping: No. With hardware raid (that doesn't suck), latency doesn't increase until you 'fill the pipe' If you're running U320 SCSI, latency doesn't increase until the combined bandwidth nears 320Mbyte/second limit of the bus itself. SATA and/or software RAID may have a bit more overhead, but you're still talking maybe 5-10% increase in latency total, not 100% per drive. Otherwise, every gamers' RAID system would boot XP/Vista *much more slowly* as it's a lot of small files loaded serially. In your scenario, RAID would extend boot times to 15+ minutes. Heck, my DEC 3000 Model 300 didn't even boot that slow! And I also quote: " SSDs have nanosecond (that's 1/100th of a millsecond) access times." Uh, your math is *way off* here also - a microsecond is 1/1000 of a millisecond, a nano-second is 1/1000 of a microsecond; in other words, a nanosecond is one millionth (1/1000000) of a millisecond. A quick google search shows that the average access time for an SSD is around 0.1 to 0.3 milliseconds... that's 100000 to 300000 nanoseconds. Do you bother reading what you write before you hit 'submit?'

"Plus, flash still seems to be somewhat less reliable than mechanical. I would rather buy a mechanical drive rated for 1.2 million hours than a flash disk rated for 100,000 operations." You guys just crack me up about this reliability issue. That's 100,000 WRITES to the SAME SECTOR. Assume 1 write per second to the SAME SECTOR (drives load balance to evenly distribute writes across the media) then the media will last about 27 days. HOWEVER the drives do write distribution which significantly increases this to way beyond the lifetime that would normally be associated even with rotational media. Who's using their drive from 5 years ago, let alone 10? Has *ANYONE* *EVER* heard of a write problem with SSD and sector wearing???? Anyone?

If bus errors don't catch you, resilvering will. When one of your disks fail, there's a really good chance you'll get unrecoverable data errors when resilvering. RAID 5 is not good enough for 8 disks anymore, RAID 6 is required. And if you're any smart, you use a real filesystem like ZFS raidz2. Using Adaptec etc. RAID adapters is just asking for trouble. With data volumes that huge, bus errors will be almost certain. You want your checksumming and parity calculations happening on the main CPU instead, preferably with some mirrored ECC memory and CPU that's at least a bit fault tolerant against random bit flips.