Unusual SSD use - recommendations?

[Stilez]

Hi,

I'm looking for recommendations for a 30 - 60 GB SSD ($60 - $125) to hold the paging file on my win7 workstation.

The machine is quad core and has high RAM use (mainly due to heavy browser use, 1GB+ Excel documents, and multiple VMware virtual machines). It handles sensitive data so the paging file uses a page file encrypting utility. Microsoft recommend that the paging file is a good candidate for SSD. However as the page file on my system will be incompressible I'm wondering whether to go for an Indilinx Barefoot SSD rather than Sandforce, and if so which one?

I'm not worried about early SSD burnout, it only has to last a couple of years. My main motive is that I've maxed out RAM on my motherboard and an SSD based page file seems one way to deal with the inability to add extra RAM. I'd like to defer upgrading the whole PC until 2012.

I'd be happy to go for another Vertex 2, but given the unusual use (incompressible data, paging file access patterns) thought I'd ask for recommendations.

Thanks!

 

[VirtualLarry]

TBH, I would go with an Intel X25-M 80GB.

The reason for the size is longevity, writes get spread out over a larger pool of flash, thus lasting longer.

 

[Stilez]

A year ago I agree, Intel would be the obvious choice. But in this case, the SSD is heavily overprovisioned anyway (30 - 60 GB for a maximum 6 GB swap file) and longevity isn't an issue as I don't expect to need it for more than 1.5 years. Raw speed for handling incompressible pagefile type use is the key.

 

[Zap]

However as the page file on my system will be incompressible I'm wondering whether to go for an Indilinx Barefoot SSD rather than Sandforce, and if so which one?

Indilinx drives are outperformed by Intel and Sandforce drives in random writes, regardless of whether or not data is compressible.

 

[LokutusofBorg]

The only thing "unusual" about your SSD usage, as you say, is that the file will be encrypted. You'll still see the typical page file IO profile, which all SSDs excel at. And the SandForce drives are still well worth buying even with incompressible data.

 

[Stilez]

Indilinx drives are outperformed by Intel and Sandforce drives in random writes, regardless of whether or not data is compressible.

That last bit is what I wasn't clear about from the various performance reviews.

Which reviews/tests/stats are the giveaways on this point?

 

[Mark R]

Hi,

I'm not worried about early SSD burnout, it only has to last a couple of years. My main motive is that I've maxed out RAM on my motherboard and an SSD based page file seems one way to deal with the inability to add extra RAM. I'd like to defer upgrading the whole PC until 2012.

You say that burnout may be not be a problem, and you are probably right.

However, indilinx drives seem to have much weaker wear management than other types. Some people have reported taking their indilinx drives to the end of life (as reported by the drive via SMART monitoring - rather than having the drive die) within a year of normal home use.

Heavy usage with small pagefile type IOs, is particularly hard on SSDs - so I would consider a controller with better wear management.

 

[Lead Butthead]

with memory so cheap these days, why would you even bother with page file?

 

[Voo]

with memory so cheap these days, why would you even bother with page file?

Because everything above 4gb sticks is far from cheap (and even those are expensive when compared to 2gb), even then we're far from the 80gb a SSD would offer and you can have both?

 

[Stilez]

However, indilinx drives seem to have much weaker wear management than other types. (...) Heavy usage with small pagefile type IOs, is particularly hard on SSDs - so I would consider a controller with better wear management.

Are you thinking of SandForce/Intel here, or some other? (The C300 is aggressive on some stats but reviews suggest the 60GB model may be a lower performer in raw speed terms and less suited for this kind of use).

Can you show me a bit more clearly what to look for in existing reviews when picking a controller by its wear management characteristics?

 

[Nothinman]

with memory so cheap these days, why would you even bother with page file?

Because Windows still pretty much requires one so running without one is stupid and the OP specifically mentions that his workload generates a lot of pagefile activity.

 

[Old Hippie]

Can you show me a bit more clearly what to look for in existing reviews when picking a controller by its wear management characteristics?

I'd like to see it too because AFAIK they're aren't any.

 

[mclaren777]

I know this is slightly off topic, but how does an Excel document grow to be 1GB+? 😱

 

[Stilez]

I know this is slightly off topic, but how does an Excel document grow to be 1GB+? 😱

To stay on-topic, can I skip that for now? I'll explain after the SSD inquiry is done. 4 word answer - "large scale commercial modeling".

 

[Mark R]

Are you thinking of SandForce/Intel here, or some other? (The C300 is aggressive on some stats but reviews suggest the 60GB model may be a lower performer in raw speed terms and less suited for this kind of use).

Can you show me a bit more clearly what to look for in existing reviews when picking a controller by its wear management characteristics?

There aren't any reviews on this topic because it's virtually impossible to test.

This is from my own experience based on statistics provided by a number of people, and which I've read on a number of forums.

My first hand experience with Intel, is that even under very heavy home usage, their anticipated life is far in excess of any reasonable expectation. Figures from people with Sandforce drives suggest that they are at least as good, if not better.

 

[Zap]

Indilinx drives are outperformed by Intel and Sandforce drives in random writes, regardless of whether or not data is compressible.

That last bit is what I wasn't clear about from the various performance reviews.

Which reviews/tests/stats are the giveaways on this point?

See the preliminary test data of the Sandforce 2000. On sequential writes, the performance of incompressible data was half compressible data. On random writes it was more like 2/3.

Let's use the worst case scenario of "half" as an assumption for the SF1000 series, since that's a good general baseline compression rate.

Look at these charts and even halving the Sandforce numbers, they still spank Indilinx in random writes, let alone aligned random writes.

There's no way to sugarcoat it. Indilinx random write performance is weaksauce. Even a 40GB Intel SSD with a "rated" 35MB/s which is a fraction of the "rated" performance of Indilinx Barefoot drives (not to be confused with Indilinx Amigo) bests the Indilinx in random writes.

What more do you want? A tree cut down with a herring?

 

[RU482]

Because Windows still pretty much requires one so running without one is stupid and the OP specifically mentions that his workload generates a lot of pagefile activity.

I must ask....if Windows has the 'option' of disabling the pagefile, then why does windows still pretty much require one?

 

[Nothinman]

I must ask....if Windows has the 'option' of disabling the pagefile, then why does windows still pretty much require one?

The option is there so you can put it on a drive other than the system drive, not so that you can disable it completely.

 

[Stilez]

See the preliminary test data of the Sandforce 2000. On sequential writes, the performance of incompressible data was half compressible data. On random writes it was more like 2/3.

Let's use the worst case scenario of "half" as an assumption for the SF1000 series, since that's a good general baseline compression rate.

Look at these charts and even halving the Sandforce numbers, they still spank Indilinx in random writes, let alone aligned random writes.

There's no way to sugarcoat it. Indilinx random write performance is weaksauce. Even a 40GB Intel SSD with a "rated" 35MB/s which is a fraction of the "rated" performance of Indilinx Barefoot drives (not to be confused with Indilinx Amigo) bests the Indilinx in random writes.

What more do you want? A tree cut down with a herring?

Thanks - useful!

The consensus seems to be that you'd expect a small hit on reads, and a 1/2 - 2/3 speed on writes, if the data is incompressible. The next page of the same review you linked matches this well, and gives results for exactly the needed accesses with compressible/incompressible data on SF-1200. It concludes:

"Read performance (...) drops to around 80% of peak (...) Sequential write speed actually takes the biggest hit of them all. At only 144.4MB/s if you're writing highly random data sequentially you'll find that the SF-1200/SF-1500 performs worse than just about any other SSD controller on the market. Only the X25-M is slower."

Looking at the Indilinx speeds (sequential/random), the SF speeds with incompressible data, and Microsoft's analysis of pagefile use in Windows, the expected speeds for an incompressible pagefile would be:

• Pagefile writes (1 in 41 accesses, large/seq) @ 144 MB/s (SF) 184-192 MB/s (Indilinx)
• Pagefile reads (40 in 41 accesses, small/random) @ 42 MB/s (SF) 35 - 39 MB/s (Indilinx)

Is this analysis reasonable? Does it still suggest SF?

 

[Zap]

I must ask....if Windows has the 'option' of disabling the pagefile, then why does windows still pretty much require one?

Windows itself doesn't need it. Some software do.

Is this analysis reasonable?

I think the time you've spent analyzing this is more than the time you will save in the next 5 years due to using a "faster" SSD over a "slower" one.

 

[Old Hippie]

I think the time you've spent analyzing this is more than the time you will save in the next 5 years due to using a "faster" SSD over a "slower" one.

+1

Just buy a Crucial C300 series and be done with it. 😀

 

[Stilez]

I think the time you've spent analyzing this is more than the time you will save in the next 5 years due to using a "faster" SSD over a "slower" one.

Part true. But it might not have been, you can only say what would be the case once informed.

The insight and links are also useful for future.

 

[Voo]

Windows itself doesn't need it. Some software do.

Yeah to elaborate on that: Beside the default use case, there's some old software that uses the pagefile in unintended ways even if you have enough RAM on the system which leads to some annoying crashes. I remember last time that topic was brought up, someone named a few things, nothing I use so didn't stick, sorry.