Use mSata to reduce writes to main SSD?

I'm looking to buy a workstation laptop (likely Lenovo W530) with a 256-512GB SSD as main storage and 500GB+ HDD in the bay.
What I'm wondering about is how to best use my mSata connector since I'm not interested in WWAN.

I'm aware that the mSata is only Sata II while the main SSD and HDD are Sata III.

So, here's my idea: how about getting a small-ish, cheap-ish mSata SSD and use it as a dedicated "garbage collector"?

In other words, reserve it to store things like browser cache, scratch disk, temp files... you get the idea.
The idea is to minimize writes to the (expensive) main SSD with only a minimal performance hit, thus extending its useful life.
Whenever the el Cheapo mSata is busted from all that "abuse", I'll just get another one... I'd rather replace a <$40 than a >$250 unit!

In this scenario I would store OS, apps, games and "important" data on the SSD, most other files on the HDD.

I'm also thinking that by NOT caching the HDD in the caddy I would be able to swap it out for an ODD any time I want without ill effects.

I haven't found any reference to such a scenario here, thoughts?
What would be the ideal size for the mSata (also considering bang for the buck), is 32GB already overkill?

 

Not worth it at all. SSDs are more likely to die from controller failure rather than running out of P/E cycles. To put things in perspective, for a 256GB SSD with 3,000 P/E cycle NAND, you'd need to write around 768TB data before you run out of P/E cycles. The average user writes around 5-10GB/day. At 10GB/day, it'd take 210 years before you'll run out of P/E cycles (course, I don't think it'll be able to hold a charge that long but you get the picture).

 

here's another idea: what if you buy a nice SSD that will be able to handle more writes than the laptops useful life, and you dont have to worry about wearing it out

I happen to have two of those, lol. Endurance testing shows such a drive writing over 5PiB (5+ million GB) and keeps going with steady average speed of about 300MB/s. Impressive is not enough to describe these drives, heh.

 

Thanks for the quick replies... I'm a bit surprised by your reactions!

Let me explain why:
My limited SSD experience is with two older models which I use to run Windows 7 Pro on two desktops.

SDD 1 is an OCZ Vertex 128GB.
I set it up to offload as many writes as possible (mostly to a HDD-based ramdisk).
After almost 3 years the drive health is at 61% with 3.54TB lifetime writes.

SDD 2 is a cheap OCZ Onyx 32GB.
I just threw the OS on it and let it do it's thing.
After only about 6 months it logged 1.54TB lifetime writes, and the health is already down to 31%!
(I use Hard Disk Sentinel Professional to monitor my drives).

So, what gives?
Do the newer generation SDD's have a lot more PE cycles, are the values reported by HDD Sentinel completely bogus, or did I mess up somewhere?

 

I've got an Intel X25-M G2 120GB with 3.55TB writes and still at 99% drive life and a Kingston SSDNow V 40GB (rebadged Intel X25-V) with 1+TB writes still at 98% (need to double-check for specific values). I think you just happened to choose really crappy drives or drives with really high write amplification (e.g. with 10x write amp, you write 1GB to the SSD and the controller writes 10GB). Newer, smaller lithography NAND actually tend to have less P/E cycles. However, newer controllers have improved a lot to reduce write amplification.

Go with something like the Samsung 830/840 Pro, Crucial m4, Plextor M3/M3P/M5S/M5P, etc, instead.

Edit: Just checked the Kingston, 1.17TB writes, 11,000 hours, 99% drive life.

 

The small capacities you buy are 'wrong' - even if your LTW are on the (very) low side.

They are 'wrong' because such low capacities do not allow you to leave any usable real world capacity for your workload AND partition them smaller to allow TRIM, GC (garbage collection) and other internal housekeeping to be carried out with the least WA (write amplification) and sustained highest performance possible with any workload you throw at it.

For example: in my heavily used systems: I leave up to 60% of the SSD as 'unallocated' (via partitioning from first use/brand new status). This results in a setup that does not slow down - no matter what workload I put on the storage subsystem. In addition; I am keeping the nand as healthy as they can be (by keeping the WA to a minimum - always).

Yes: this is not a misprint: I partition a 240/250/256GB SSD as ~100GB for O/S and Program use and leave the rest 'unallocated'. And; this is different than simply leaving free space on the drive, if you want to minimize WA while also keeping the sustained performance as close to 'new' as possible.

I also only use the most efficient capacity SSD's too: 240/250/256/512GB (depending on the controller used).

Anything smaller and the controllers are not fully utilized (for performance and longevity; all controller channels populated, interleaving nand and extra bonuses (firmware/hardware) that further differentiates larger capacity drives...) and the sheer capacity of the drive also helps keep the drive much healthier (with more nand; more LTW can be achieved; even with the same workload that would 'fit' in a smaller capacity drive).


To answer your specific question:

Yes, I would do an mSATA boot drive - a 240/256GB (Crucial M4 mSATA) and partition it to ~100GB and never worry about it again.

With the remaining drive bays: I would be using a Crucial M4 512GB for my 'Work In Progress' drive (and partitioned to around 400GB or smaller) and the final drive bay would be either another 512GB M4 (similarly partitioned) or, a 500GB 7200 RPM drive to backup the WIP SSD. (Why not a larger mechanical drive? Because denser capacities make a HDD slower: not faster).

With Win8; set up the File History (similar to Time Machine in OS/x) to use the backup drive and forget about backing up again (tip: make sure you include the WIP drive as a 'Library' location - then it will be automatically backed up too).


With the above setup, along with Win8PROx64 and at least 16GB RAM, I would not be worrying about HDD Sentinel reporting any guessed values properly or not (nor would I ever care to even see them reported properly...); nor would I ever consider using an SSD as a cache drive - or such a tiny (capacity) SSD ever again period.

Yeah; the performance difference is that great. (Especially sustained performance over time).

Hope this helps.

Good luck.

 

Both your drives use older Indilinx controllers, which have pretty terrible write amplification. Make sure you are on latest firmware because older firmware is even worse. This article explains what write amplification is, but basically whatever you write to the drive, the drive ends up having to write MORE. Modern (AKA the current decent drives on the market) do an okay job in keeping it between 1 and 2 on average (Sandforce can do under 1 due to data deduplication). Older drives were terrible in this respect. I can't find any sources right now, but I recall hearing of write amplification between 10 and 100 for older drives, especially Indilinx.

A few of the replies have some good information.

You see that your 32GB drive used up its life pretty quickly.

All that said, at the current rate your Vertex will last nearly a decade. Are you still all that worried about it? How many decade old computer parts are you still using today?

 

Wow some really thoughtful replies here, thanks a bunch!

@tilleroftheearth: very very interesting reply, I will mull that over.
Are you saying that I should put Windows 7 entirely on the mSata (despite only sata II), or only the boot partition (directing for example \Program Files to the main SSD)? I know that with SSD prices coming down it's no longer such an issue, but wasting nearly half of the capacity on essentially damage control (by overprovisioning) still seems painful to me...

@zippyzap: Yeah I know my SSDs are old, small and crummy. Hey, they were expensive back then! (at least the Vertex was).
Didn't realize how terrible their WA was, thanks for the link.

Of course I'm not worried about my Vertex. What prompted me to start this whole thread is that I believe my Vertex is only lasting so long because I'm trying so hard to avoid writes (for instance my whole \User directory is hardlinked to a HDD). It's the Onyx that really caused me to worry since I can almost watch it die even though it sees only very moderate use.

But anyway, now that I know about the advances in WA I'm much relieved!

 

Does the latest W530 only have a SATA2 mSATA connection? I would think it is SATA3?

Either way; yes I would put the O/S and most programs on the mSATA - no re-directing of any Windows 'core' directories needed.

With any heavy duty program; you can always point the install to the SATA3 SSD's...

This seemingly 'excessive' overprovisioning is not painful: what is painful is paying top dollars for a setup (like a workstation... wink wink) and getting spectacularly less than HDD performance from it in a few weeks if you don't.

Your call.

I prefer to have the highest performing setup possible - even if it means I am paying ~double the effective $$/GB ratio of what is currently considered 'normal'.

 

I don't think it needs to be as drastic as 50% OP. Even an additional 20% should do wonders.
Intel SSD 520 Enterprise Review

Keep in mind, your Onyx only has 32GB NAND. Your Vertex has 4x that amount. For obvious reasons, the Vertex would be able to withstand more writes compared to the Onyx.

 

I agree that it doesn't have to be as drastic as 50% (or more) - but when I notice performance degradation on SSD's with just 30% filled (just installing the O/S and programs), why not?

They're just too cheap to worry about (for the performance they offer) anymore.

(Says the user that considered $600+ Raptors and $400+ Hitachi TravelStar 60GB HDD's as 'bargains' so few years ago - OMG! ...2003... not 'so few years' eh?).

 

Sandforce claims lower than 1 write amplifications; guess its the only one which can claim that

 

Lower than 1 WA factor with restrictions:

Normal workload assumed (meaning mostly highly compressible data).

Low 4K R r/w demands, again; a 'workstation' type load - not a 'server' type load.

Low % filled usage (ideally; less than 50%).

At least as much idle time as 'work' time on the SSD in question (for GC) - certainly not a continuous work duty cycle.


And all the above with less than advertised raw speeds - even on 'recommended' workloads.

Yeah, SF is the only one that can claim that.



(Still love my Intel 520 Series and Sandisk Extreme's 240GB's for O/S use though, with ~60% capacity left 'unallocated').

 

The SanDisk Extreme is pretty great value for the money. I have a SanDisk Extreme 240GB for one of my VM drives (bought for $160 back when other brands were all $200+) and even with NTFS compression enabled on the virtual drives, WA is surprisingly low. Right now, it's at 438GB NAND writes and 775GB host writes for a WA factor of just 0.57x. I have it formatted to 200GiB (for 11% OP on top of the default) but so far, filled capacity has been between 90-110GiB.

 

Were Raptors really that expensive? I remember getting the RaptorX 150GB for around $180 on sale, and I think they launched at $300 MSRP. Got 300GB VelociRaptors for around $260 (launched at $300 MSRP). But yeah, funny how people complain about $200 for a 256GB SSD that is a world faster than a VelociRaptor.