There are leaked Intel SSD roadmaps on the web that talk about "Lyndonville". According to this website, this may use enterprise-MLC Nand to dramatically increase the Nand flash lifetime. This may be ideal for your usage model:
Does TRIM work correctly on an OS software raid
whose members are set to AHCI mode e.g. for dedicated data partition(s)?
As a general rule, should MLC SSDs be assembled in RAID 0 arrays,
in order to spread the wear?
In other words, is this theoretically correct and can this be confirmed experimentally ...
50% reduction per drive w/ 2 x SSDs in RAID 0? (half as much per drive)
67% reduction per drive w/ 3 x SSDs in RAID 0? (one-third as much per drive)
75% reduction per drive w/ 4 x SSDs in RAID 0? (one-fourth as much per drive)
Lastly, has Intel done any research using SSDs with Windows NTFS compression enabled?
(in drive letter Properties, see: "Compress drive to save disk space" and a check box)
1 of 1 people found this helpful
The more data written to a SSD, the more wear occurs regardless of RAID or not. So yes, there is an about 50% reduction in writes on a single member of a 2xRAID0.
As for NTFS Compression: http://blogs.msdn.com/b/e7/archive/2009/05/05/support-and-q-a-for-solid-state-drives-and.aspx
Is NTFS Compression of Files and Directories recommended on SSDs?
Compressing files help save space, but the effort of compressing and decompressing requires extra CPU cycles and therefore power on mobile systems. That said, for infrequently modified directories and files, compression is a fine way to conserve valuable SSD space and can be a good tradeoff if space is truly a premium.
We do not, however, recommend compressing files or directories that will be written to with great frequency. Your Documents directory and files are likely to be fine, but temporary internet directories or mail folder directories aren’t such a good idea because they get large number of file writes in bursts.
Thanks! Based on those graph in the frist post link, I would conclude that it is still not suitable yet for my research data to be stored there. However maybe instead of tackling the problem from the SSD side, I can tackle the problem from writing less.
Thanks all for your answers, you've all been very helpful to me!
My office is also responsible for maintaining a research database.
Here's what we have found:
(1) a 2xRAID0 of rotating platters is generally faster, and
contrary to widespread public opinion, the load on each is
roughly 50% over time;
(2) if one member fails, the practical effect is the same
as a single drive failing: the failed drive must be replaced and
the data restored --> "same difference" practically speaking
(3) quality input power is a must, e.g. good UPS;
(4) proper environmental control is also a must:
keeping dust filters clean, keeping ambient temps
under control and within range, etc.
(5) all high-performance HDDs must be installed
in drive cages with active cooling;
(6) the HDDs with 5-year warranties are more reliable,
and the cost per warranty year is usually preferred
as compared to 3-year warranties; e.g. Western Digital
RAID Edition HDDs ("RE");
(7) a key management policy is to EXPECT a failure,
and to have enough backups to recover gracefully;
(8) application of rigorous probability and statistics
urges this conclusion: the conditional probability
that both drives fail, given that one drive of a 2xRAID0 fails,
is extremely small; and, 50% less WRITEs are a major
factor increasing longevity;
(9) data rate is directly proportional to track circumference
on all rotating HDDs, because they maintain the same or
similar recording density on all tracks; thus, more frequently
accessed files should be allocated to "short-stroked" regions
on all HDDs, if possible;
(10) short-stroked drives move the READ/WRITE armature less,
which in turn reduces wear on the armature bearing and
servo-mechanism; and, in combination with RAID-0 arrays,
short-stroking is enhanced in proportion to the number of
RAID-0 members (a 100GB RAID0 partition uses only
the first 25GB worth of tracks on a 4x RAID0);
(11) for temporary files like browser caches and OS swap files,
consider a generous ramdisk: this offloads rotating platters,
further reducing wear on the latter; and, properly configured,
ramdisks can increase real-time performance 15- to 20- TIMES
over rotating platters; cf. my review of RamDisk Plus here:
(DDR3 has become even faster, since that review was written:
and, quad-channel memory access is expected for LGA2011 sockets)
(12) we're still waiting for an answer to our question about TRIM
on OS software RAIDs when members are configured in AHCI mode;
(13) all of the above strongly suggests that the Media Wearout Indicator ("MWI")
on modern Nand Flash SSDS will decline half as fast with a 2xRAID0,
one third as fast with a 3xRAID0, and one-fourth as fast with a 4xRAID0,
simply because of the way "striping" happens on all RAID 0 arrays;
(14) an experiment that Intel should consider doing, and publishing,
is to formulate a proper experimental matrix, and test the real effects
that result from enabling NTFS compression on Nand Flash SSDs,
with and without RAID configurations; SandForce controllers do
real-time compression, reportedly to reduce WRITEs as much as possible;
why not make this the DEFAULT that is performed by the OS, regardless
of the SSD's make or model?
(15) the results of proper experimentation at (14) above should
produce data that will suggest an optimal policy: for example,
perhaps a 3xRAID0 does more to reduce WRITEs to each SSD member
than default NTFS compression, due to the type of compression that
is done in real-time by Windows operating systems (LZ77?):
(16) and, given the proliferation of multi-core CPUs, in addition to those
that also support Hyper-Threading, there should be otherwise unused
CPU core resources to perform decompression rapidly in real-time, without
adding too much overhead or causing excessive additional latency
when accessing compressed files; SpeedStep and Turbo Boost
will only help mitigate this overhead, if they are activated to help
accelerate real-time decompression.
Those are just some lessons we've learned here, after many years
at Hard Knocks University.
Hope this helps.
/s/ Paul A. Mitchell, Instructor,
Inventor and Systems Development Consultant
All Rights Reserved without Prejudice