News - SanDisk claims SSD innovation improves performance and reliability

[HEXUS]

New technology called ExtremeFFS improves SSD speed and endurance according to Sandisk?s new metrics.

Read more.

[Steve]

From my perspective (which needn't be at all correct) the SSD performance issue is this:

SSD's use the same interfaces and logical block addressing that we've been using on hard disks for years. A hard disk breaks down into cylinders, head and sectors. The addressing exploits this known arrangement with things such as tagged command queuing and native command queuing.

With flash technology - unless somebody changed the way it works and failed to tell me - you can't just write to a 'sector' (or block). A chunk of the flash must be erased before it can be reprogrammed - this is often larger than the block being written. If the chunk already has data in it, the stuff that isn't being overwritten will have to be read out before the erase, then written back in.

So you can probably see why writing little bits of data to random blocks on an SSD isn't great. SSD can have incredible sequential throughput, so data transactions (writes in particular) should be twiddled with to make the most of that. The same is true for hard disks, of course, but they have mechanical limitations, meaning SSD's can create new ways of dealing with 'the problem'.

I love SSD, and folks like Intel and SanDisk are playing a big part in bringing it to the mainstream.

In a couple of years I expect we'll see crazy storage subsystems that seamlessly combine HDDs and SSDs to achieve optimal power consumption and gigantic capacities.

[kalniel]

Virtual RPM as a speed measurement sounds absolutely rubbish - are we also going to get virtual platter size/density?

A better longevity measure would be very welcome though - how about a simple 'Best before'?

[shaithis]

In a couple of years I expect we'll see crazy storage subsystems that seamlessly combine HDDs and SSDs to achieve optimal power consumption and gigantic capacities.

In a couple of years I would love to see SSD completely replace HDD in the same way as LCD has replaced CRT.

It may take a bit longer but I think it will happen sooner then a lot of people expect. The new technology has so many plus points over the old technology that replacement will happen at a phenomenal rate.

[miniyazz]

Optimisation for Windows 7, if it brings real benefits, will be one thing encouraging me to make the switch from Vista.
Also, if the essential Vista architecture remains in Windows 7 (despite being refined somewhat), is this likely to mean most Vista programs will work on 7?

[dangel]

Optimisation for Windows 7, if it brings real benefits, will be one thing encouraging me to make the switch from Vista.
Also, if the essential Vista architecture remains in Windows 7 (despite being refined somewhat), is this likely to mean most Vista programs will work on 7?

Yes, the driver architecture and API is much the same. From initial tests all my code works without (any) changes. TBH problematic programs from the XP era remain problematic by design, rather than due to OS changes (i.e. bad code = problems regardless).

[shbris]

In a couple of years I expect we'll see crazy storage subsystems that seamlessly combine HDDs and SSDs to achieve optimal power consumption and gigantic capacities.

hybrid drives anyone? dont think they ever worked that well together, but maybe in a few years they'll perfect it

[TheAnimus]

Optimisation for Windows 7, if it brings real benefits, will be one thing encouraging me to make the switch from Vista.
Also, if the essential Vista architecture remains in Windows 7 (despite being refined somewhat), is this likely to mean most Vista programs will work on 7?

7 is apparently going to be further optomised. Vista is, XP is, 2K is.

Embedded has the EWF which is a really good performance booster too!

[aidanjt]

Virtual RPM as a speed measurement sounds absolutely rubbish - are we also going to get virtual platter size/density?

A better longevity measure would be very welcome though - how about a simple 'Best before'?

The best performance metric is still internal drive media throughput. It doesn't matter how fast the system interface is or the rate of rotation of hdds, it all depends on the data rate performance between the media and the SATA chip.

So yes, 'Virtual RPM' is utterly useless in every respect.

[b0redom]

In a couple of years I expect we'll see crazy storage subsystems that seamlessly combine HDDs and SSDs to achieve optimal power consumption and gigantic capacities.

I was at a briefing by Sun on Thursday. They are planning on launching this in the very near future. The idea is that ZFS will use SSDs for caching and backend SATA drives for bulk storage. Looked very cool.

With flash technology - unless somebody changed the way it works and failed to tell me - you can't just write to a 'sector' (or block). A chunk of the flash must be erased before it can be reprogrammed....

Not necessarily true - it depends on the FS. There's (as far as I can see) no reason why you can't do a copy-on-write model FS and have the OS expire and clean the old blocks in the background. As long as the OS keeps a handle on the 'blank' sectors what is to stop it just overwriting them?

[Steve]

I was at a briefing by Sun on Thursday. They are planning on launching this in the very near future. The idea is that ZFS will use SSDs for caching and backend SATA drives for bulk storage. Looked very cool.

Awesome. ZFS looks pretty kickarse.

Not necessarily true - it depends on the FS. There's (as far as I can see) no reason why you can't do a copy-on-write model FS and have the OS expire and clean the old blocks in the background. As long as the OS keeps a handle on the 'blank' sectors what is to stop it just overwriting them?

I guess so, but you're talking about the FS, whereas I was talking about down at block level on the device itself. For that to work you have to use an FS aware of the caveats of the underlying device - and filesystems traditionally favour CHS based hard disks.

[b0redom]

Yeah, I really like ZFS - I hear that Sun are talking about GPLing it which would make mass market adoption much much quicker. It's all pretty academic as far as most end users go as I can't see it being incorporated into an MS operating system - not unless they licensed it on BSD terms anyway.

[Shooty*]

Whoa, they've called it Extreme For F**k's Sake?


That roughly approximates the sounds I make trying to install an unformatted drive.

What's next? The "UltraMEH" partition system?

[aidanjt]

I guess so, but you're talking about the FS, whereas I was talking about down at block level on the device itself. For that to work you have to use an FS aware of the caveats of the underlying device - and filesystems traditionally favour CHS based hard disks.

The thing is though, filesystems are in a better position to assess what is going on with the data. And filesystems like Reiserfs and XFS already operate in blocks, and allow you to place the journal in an external block device.

[TheAnimus]

there is a lot to be said for getting the actual device to do the 'drive device specific' style optomisations, the pace at which we've seen SSDs performance patterns change, this would be prudent.

A technology like the original EWF (which is about 6 years old iirc) is pretty much useless, why should the OS have a wear reducing write pattern, when only the drive truely knows where its most likely to be warn (ie has it been used in a difference PC, but also some areas are more robust than others.....).

[Steve]

there is a lot to be said for getting the actual device to do the 'drive device specific' style optimisations, the pace at which we've seen SSDs performance patterns change, this would be prudent.

A technology like the original EWF (which is about 6 years old iirc) is pretty much useless, why should the OS have a wear reducing write pattern, when only the drive truely knows where its most likely to be warn (ie has it been used in a difference PC, but also some areas are more robust than others.....).

True. At the block level the file system should really only need to make a few assumptions. Such as: If the device access time isn't constant across the address range, stuff put at the beginning should be quicker to get at - and maybe: sequential transactions are better than random ones.

The rest should be up to the device firmware.