SanDisk thrusts SSD into the client OEM battlefield
Taking on Hitachi, Intel, Plextor and the gang
Cloud storage: Lower cost and increase uptime
SanDisk has dived into the client OEM flash drive jungle with its latest X100 SSD.
SanDisk bought SSD controller company Pliant last year and that company's Lightning SSD products provide SanDisk with its enterprise server-class products. Now it has introduced its X100 line of client SSDs – PC, notebook and ultrabook – which do not appear to use Pliant controllers.
We have obtained fuller details of the device and its speeds and feeds are:
- 32GB, 64GB, 128GB, 256GB and 512GB capacity points;
- 24nm, 2-bit MLC NAND from, we assume, Toshiba;
- Random read/write IOPS (4K blocks) are 38,000 and 22,000;
- Sequential read/write bandwidth of 500MB/sec and 420MB/sec;
- 6Gbit/s SATA interface for 2.5-inch form factor;
- 2 million hours MTBF; and
- mSATA and customisable other form factors for ultrabook devices.
The X100 features include a 3-tier hierarchical storage structure of volatile cache (DRAM is El Reg's guess), SanDisk's non-volatile nCache and bog-standard flash. The idea is that the nCache improves random write performance, by storing written 4K blocks – which are typically written in intermittent bursts – in a dedicated single level cell (SLC) region of the flash, assuming the same architecture as the existing U100 tablet SSD.
SanDisk says the nCache can be emptied in less busy times, and asserts: "For a typical everyday use, the write performance that the users see is the nCache (burst) high performance, and not steady state (sustained) ... performance." In other words, random write performance would worsen if the nCache filled up.
Overall: "Data pattern streams are ... monitored and rearranged by a proprietary innovative multi-streaming feature that reduces fragmentation and improves locality of data. This enables fast user response, no stuttering, better multitasking capabilities and significantly improves the drive’s long-term data endurance6, ensuring an enhanced user experience."
A patrol read mechanism does background read checks with error correction. The X100 can adjust its performance to fit within different power classes or envelopes.
SanDisk says the X100, with its TRIM support, can endure 80TB being written (TBW) to it, assuming what it calls a 128GB typical Windows 7 workload. The company says: "Data is written using typical PC transfer size, written at a constant rate over the life of the SSD, and data is retained for at least 1 year upon TBW exhaustion. Based on SanDisk internal measurements, a typical client PC user writes 4GB/day."
The X100 has a 2 million hours mean time between failure rating.
Okay, so how does the X100 compare to some other client MLC SSDs that OEMs could choose?
- Hitachi GST's Ultrastar SSD400M does 56,000/24,000 random read/write IOPS – significantly better; 495/385MB/sec sequential read writes – a little worse; and 7.3PB lifetime writes – significantly better.
- Intel's 520 does 50,000/42,000 random read/write IOPS – significantly better; 550/520MB/sec seq read/writes – rather better, but we don't know its endurance, although there's a 1.2 million hours MBTF – which is not so good.
- Plextor's M3 does 70,000/65,000 random read/write IOPS (wow!); 525/445MB/sec seq read/writes – a bit better; and has a 5-year warranty as well as claiming a lowest annual failure rate of 0.05 per cent.
Based on these spec sheet numbers, the X100 is a long way from being an IOPS leader, or a MB/sec leader – although it has good sequential bandwidth numbers – and isn't an endurance leader in TBW terms. It appears the nCache feature is not delivering that significant a boost to random write speed and the X100's random IOPS numbers don't shine in this company.
SanDisk must think that when potential OEMs make more detailed comparisons, the X100's mix of features at the various capacity points and prices will win the day. ®
COMMENTS
Re: The reliability of SSD devices still worries me
"""Suffice to say, SSDs haven't proven themselves both over the long term and at high storage densities."""
They might not have done so for you, but I've got hundreds of them that've been running 24/7 in enterprise workloads for the last 3 years... they're fine.
"Moreover, figures for read/rewrite operations are actually published, this was never 'matter of fact' for magnetic HDs; alone, this is of considerable concern (especially when considering long term storage)."
I think maybe you don't understand how SSDs work, and specifically how they're different from spinning magnetic storage devices. Also, manufacturer-published performance data is always worthless, on any product in any field. Test it yourself. Long term storage, which to me means "Write once and read the same data over a long period of time" is a perfect application for an SSD - they're really quite good at maintaining written data, since most of the failure modes occur on write events, not reads. A spinning disk, on the other hand, is more or less guaranteed to fail in 3-5 years of continuous runtime.
And you shouldn't have any data, on any storage device, without backups (and RAID, if space allows.) If you follow those same standard guidelines with SSDs, you still won't have data loss.
Just how do they calculate the mean time between failures?
They quote 2 million hours (about 200 years), so do they test 1000 of them for a few months and extrapolate from that? Or do they just pluck a figure out of air at random?
Paris because I'm sure she could go for 2 million hours :)
The reliability of SSD devices still worries me
Today, I've already had a long rave about SSD storage devices under the title of 'Intel 520 240GB SSD' and there's no point repeating all of it here.
Suffice to say, SSDs haven't proven themselves both over the long term and at high storage densities. Moreover, figures for read/rewrite operations are actually published, this was never 'matter of fact' for magnetic HDs; alone, this is of considerable concern (especially when considering long term storage).
Quoting from earlier remarks: "...it remains to be seen how they work out in practice, certainly for the moment I only use them for OSes or where backups are guaranteed and regular."
IT infrastructure monitoring strategies
What you need to know about cloud backup
Enabling efficient data center monitoring
Agentless Backup is Not a Myth
Top 10 SIEM implementer’s checklist
