Intel X25-E Extreme SSD Benchmarks On Linux

[paul_one]

I really didn't like the fact that the tests were purely CPU intensive either.

Are the reviews really headed towards Toms Hardware grade quality?

How about moving files? Downloading files? and reading/writing tasks - which are the whole appeal of these drives?

Can you move 3 files around different partitions and send files over the network AND to a USB drive at a good speed?

How long does it take to start up to the bootscreen (bootchart?), how about starting/stopping common programs/services?
How about starting many different programs at the same time - how does this effect times?
What about watching a movie, listening to ogg files, ripping files, and several low CPU-intensive but many operations/file based ops - which is CERTAINLY real world for a LOT of people.

... How about battery life with these tests too?


Along the issue of cache - I'm thinking he means FILESYSTEM cache, which is written to RAM before being flushed to the disc using fsync/etc. This can be different depending on workloads/RAM/etc.
A time-line graph (I'm just thinking performance against time - don't know the proper name) would probably do a better job at monitoring speeds but I guess the performance stats were average-based and didn't have any hooks into the kernel/whatever.

... Can't some stats be gained from /proc/ at intervals? (for those in the know)


Overall, my criticism on this article (and only this, as phoronix seems to have written a few very nice and interesting articles - and this annoys me enough to join and post about it) is that the article takes two drives, and want's to test any performance difference between them (real world or not)... BUT the tests which are performed are totally inadequate for the initial proposal, and actually test the wrong area of the laptop altogether (the CPU).
... You may as well have just left it running in the corner of the room, or given it to Joe from the coffee shop on the corner and asked "do YOU think it's faster?".

Please phoronix, please get people who THINK about the tests they perform, and HOW to go about them (perhaps ask in the forums for ideas?) rather then people who just run tests at the time and decide they're relevant.

[bexamous]

You can use iostat to get data to graph maybe?

but yeah not sure about these test results because that IOZone read at 350MB/sec is impossible. SATA2 is only 300MB/sec :P. Need to drop caches before running these benchmarks.

Also whatever you want to regard as a 'good' benchmark... maybe try running it in succession for awhile to see if performance degrades due to SSD.

[kxmas]

I agree with most the comments on here. What a completely worthless article.

How about showing boot times, application launch times, and some sort of productivity suite test?

[Svartalf]

Too expensive for normal people who need big storage capacity. I wonder in how many years SSDs will drop to the price of today's HDs and totally replace them. Probably too many :P

Not as many as one would think if Micron's actually pulling together their line like the word going around would indicate. We won't know until a couple of months from now whether that's the case or not, but it is worth noting that the drive Michael Larabel reviewed here is an SLC drive and as such is going to be smaller and much, much more expensive. $550 is the rough price for a slightly slower SLC NEBS Level 3 certifiable 64Gb drive from Samsung right at the moment- we priced them out and I almost got product management to sign off on them this go round for our replacement for magnetic disks.

[Svartalf]

More to the point, you made no mention of the common issues surrounding SLC Flash SSDs: re-write speed. The performance of single-level cell flash, when re-writing, is hideous.[/URL]

Actually in a shoot out between the current NEBS certifiable (Telecom grade...needs to be about as abusable as the mil-spec stuff...) drives from Samsung, Intel, and the magnetic drives we have in hand- the rewrite performance wasn't anywhere near as hideous as everyone'd been led to believe. I can't share the details (Tektronix confidential info...sigh...) but we'd almost had upper management and product management sold on this generation. Still just a bit too expensive yet because they don't believe in the reliability story- they're still hung on the "disk" being in the name and that means the same thing as the magnetics in their mind.

[lem79]

I think the IOzone read performance numbers are wrong. Since when does a 7200rpm Seagate do 133mb/sec read? My 10000rpm Raptor only does about 80.

And, what's the limit for SATA2.0 interface, 3 gigabits/s? That's about 384mb/sec I guess. That SSD is coming close to maxing out 3Gbit/s.

What I think is going on is read caching (in the host's RAM, courtesy of the kernel).


Edit: I was referring to the 4Gb tests. The 8Gb read tests show something much closer to reality. Read caching for sure.

[Nexus6]

Ted T'so has a couple of great articles on SSD and filesystem configuration

http://thunk.org/tytso/blog/2009/02/20/aligning-filesystems-to-an-ssds-erase-block-size/

http://thunk.org/tytso/blog/2009/02/22/should-filesystems-be-optimized-for-ssds/

It would be interesting to see if the aligning of the filesystem to the SSD parms makes a significant difference.

Cheers,
Toby Haynes