Page file and indexing on an SSD, just how bad is it?

[[XC] Oj101]

So today I received my first decent SSD (I won't count any of the OCZs I've had - baaaad experiences there ), an Intel 320 series 120GB. Just how bad is it to have the page file on the SSD? Considering the five year warranty, is it worth the risk? If it isn't likely to kill the drive within the next three years it isn't an issue as I'll upgrade before then anyway. And indexing service? Surely when doing a search it's better to just go through the index quickly than essentially re-index the entire drive?

Thanks guys

Edit: Most of these tweaks were to prolong SSDs based on the buggy JMicron controllers, weren't they? In other words a fresh install with prefetch/superfetch disabled should be fine?

[Vapor]

Extremely, extremely unlikely you'll wear the NAND down to the point of failure. The NAND in the Intel 320s is rated for 5000 P/E cycles, has a near 1.00x write amplification, and the 320 also has a parity scheme with spare NAND. NAND is rated for a minimum of 586TiB host writes over the course of its lifetime, that's an absurdly huge number.

Long story short, you can write the full size of the drive (120GB) nearly 3 times a day, every day for the warranty period (5 years) before you wear the NAND its minimum wear rating...past that it's also extremely likely the NAND will continue to survive because A) it's good NAND, and B) the parity scheme allows for a lot of NAND failure.

Pagefile and indexing won't write nearly that much (you'll struggle to write a tenth of the NAND's rating) and having them turned on using an SSD will provide the best search performance possible and the best pagefile performance possible.

[[XC] Oj101]

Thanks Vapor, I've been following that thread but it's a bit beyond me I don't understand the write amplification at all, but from what you've said I'll be leaving indexing on and the page file on the SSD. I only have 4GB of RAM at the moment (I'm not upgrading until either 8GB sticks become popular or DDR4 arrives) and between Opera (often with 100+ tabs open, I kid you not I've had Opera using close to 2GB of RAM), WCG and GPUGrid I'm hardly ever left with a snappy machine, so having the page file on the SSD should show nicely for me

[Vapor]

Sounds like you'll write more to the SSD than the average Joe with so much reliance on the pagefile (the boost in performance vs. a HDD for pagefile usage is all the more reason to use an SSD, though), but still extremely unlikely you'll even use 30% of your NAND minimum rated lifetime over the 5 years (based on what I've seen from other power users usage rates and how good the 320s are at not wearing themselves out).

[[XC] Oj101]

Well to be honest I only put heavy use on my PC for a few hours each night (which is when I need the performance), during the day there should be minimal amounts of page file usage. I should be upgrading RAM within the next six months anyway, so I reckon I'll be fine. Thanks Vapor

[Computurd]

also, the read/write ratio on a pagefile is 40:1.
40 reads per 1 write.

*note* i havent used a pagefile for years. thats another topic.

[zalbard]

I don't think you need indexing on an SSD, all searches will be lightning fast, anyway.

[Computurd]

yes, but instead of disabling indexing entirely, just activate it only on spinners. You can move the index file to another drive as well if you choose.

[[XC] Oj101]

So after getting this Intel I haven't been as aware (read not at all) of memory usage thanks to the speedy pagefile and no longer close Opera before opening a game. Thanks to my tab addiction (I often have 100 tabs or more open spread across four windows) my "idle" memory usage is 96% with distributed computing projects using about 700mb and Opera using just under 2gb. My pagefile must be taking some heavy abuse, how much difference would upping my memory from the current 4gb of Hypers to 8gb of el cheapo memory make to the longevity of my SSD? I would imagine that the distributed computing projects also thrash the thing, but I can't really monitor it. Do they do much damage? Considering my pagefile usage I guess I'm a very user of the drive.

[Computurd]

based upon what we are seeing in the Endurance thread in this subforum, i would think you really have nothing to worry about. It would be very safe to say that you could write 20+ GB a day for five years with no ill effects.

the intel 320 that is in testing is at 437.5TB of writes so far. Not to mention it is still running and is only 40GB, so yours will have much a much higher wear tolerance.

just use it to hell would be my advice!

[[XC] Oj101]

Perfect, thanks

[uOpt]

I don't think you'll kill it but it seems quite pointless.

If your system swaps a lot (and that means actively using swapspace I/O, not just parking some bloatware there that is never used), then you need more RAM. More RAM will beat the SSD by several orders of magnitude.

If you don't swap, then the SSD is useless and wasted in that position.

Don't forget that swapout usually goes along with regular I/O. E.g. you start a new program that is read or does read blocks from a filesystem and the OS kicks out older things from RAM. Even if use swap to a SSD, the whole I/O subsystem is still double loaded. The fast SSD for swap might even slow down other disk transfers due to resource competition in hardware or software.

[sergiu]

Modern operating systems tend to swap first program data (the one that is not modified during the entire lifetime, like constants) and program code. This kind of usage will translate in a write once, read anytime scenario, as when OS needs more memory, it just delete what it already has on disk. There is also the "normal" swapping of data which might be read, modified and then swapped back. For this to be the dominant scenario, you would need to run at least two programs which are active all the time and their cumulative memory usage is well beyond total memory. If this is happening all the time, you will notice it quite easily, as even with a SSD, the throughput of the programs will be very low compared to their potential maximum. For this only more memory helps.

[[XC] Oj101]

Well as I said, before opening Need For Speed my memory usage was already 96%. NFS used about 1.1GB according to Task Manager, but the memory usage stayed at 96% (I can't seem to get 97% even if I try?). I'm reckoning that there is a lot of swapping happening, but even so it's magnitudes faster than my old shortstroked RAID0 array with memory usage in the low 20% range.

The speed is not the question at all, but wear and tear. As I still have a few drives in RAID, I can't access info such as the total amount of data written to the SSD (even Intel's utility doesn't allow anything to be done with the SSD) so I can't monitor if I'm actually doing anywhere near a "lot" of writing or if it's a few GB per day.

I think I'll wait till RAM drops to $30/4GB stick (locally) or for DDR4 before going to 16GB - and the only reason for the upgrade will be to have everything running optimally

[uOpt]

Modern operating systems tend to swap first program data (the one that is not modified during the entire lifetime, like constants) and program code. This kind of usage will translate in a write once, read anytime scenario, as when OS needs more memory, it just delete what it already has on disk. There is also the "normal" swapping of data which might be read, modified and then swapped back. For this to be the dominant scenario, you would need to run at least two programs which are active all the time and their cumulative memory usage is well beyond total memory. If this is happening all the time, you will notice it quite easily, as even with a SSD, the throughput of the programs will be very low compared to their potential maximum. For this only more memory helps.

Regular compiled code ("program data") is not moved to the swapfile at all. It is readonly mapped and if evicted to make space it is simply dropped from memory. If needed again it is fetched from the original place in the file. This applies to both exe files and dll files, plus any data that is mapped readonly, which is often the case for game data such as textures.

Swapspace is never involved here. This is another reason why swapping to SSDs is overrated. If you were to redirect the SSD money into more RAM you wouldn't have to worry about any of this.

[sergiu]

Regular compiled code ("program data") is not moved to the swapfile at all. It is readonly mapped and if evicted to make space it is simply dropped from memory. If needed again it is fetched from the original place in the file. This applies to both exe files and dll files, plus any data that is mapped readonly, which is often the case for game data such as textures.

Swapspace is never involved here. This is another reason why swapping to SSDs is overrated. If you were to redirect the SSD money into more RAM you wouldn't have to worry about any of this.

This is still a form of "swapping". Now, if dropped out of memory, it still uses virtual address space, it will still generate a page fault and a reload. This is actually a scenario where a SSD shines because of the very small read latency, as this can happen even with swapping disabled (easily observed if one application is minimized for one day). I was however on the impression that some operating systems like UNIX for example keep also libraries swapped as it is possible to drop them out partially... must study more on subject...

[uOpt]

This is still a form of "swapping". Now, if dropped out of memory, it still uses virtual address space, it will still generate a page fault and a reload. This is actually a scenario where a SSD shines because of the very small read latency, as this can happen even with swapping disabled (easily observed if one application is minimized for one day). I was however on the impression that some operating systems like UNIX for example keep also libraries swapped as it is possible to drop them out partially... must study more on subject...

The SSD used for swapspace is not involved, ever, in this scenario. The reload of a dropped page happens from the original location in the filesystem.

Neither Windows nor Unixes treat binary executables and shared libraries any different. They are simply subject to the normal paging system.

[sergiu]

The SSD used for swapspace is not involved, ever, in this scenario. The reload of a dropped page happens from the original location in the filesystem.

Original location is usually also on a SSD unless the drive is used only for swapping which is highly unlikely for a normal user. Actually this is where I saw a big difference. After switching to a SSD, I never experienced a 1-3 seconds delay at maximizing an application after 12+ hours of idle.

[[XC] Oj101]

I'm not so sure of that. If I Alt Tab out of Need For Speed to Opera and Alt F4 kill it everything happens instantly. Before, this would be at least a two minute process if the memory was full, and a good ten seconds with the memory empty. My computer boots faster, my games load quicker (confirmed with a stop watch, it's not a placebo effect) and swapping between open applications is always instant. If I move the pagefile to my R0 hard drive array, things slow down again. SSD swapping definitely isn't overrated, and the overall experience using my computer shows that no amount of memory would give me the same kind of speed boost.

[Computurd]

the page file has a mix of 40 reads to 1 write according to microsoft.

[uOpt]

Original location is usually also on a SSD unless the drive is used only for swapping which is highly unlikely for a normal user. Actually this is where I saw a big difference. After switching to a SSD, I never experienced a 1-3 seconds delay at maximizing an application after 12+ hours of idle.

Except this thread is about swap on SSD.

[bluestang]

@uOpt, so what SSD do you own?

[uOpt]

@uOpt, so what SSD do you own?

An original 30 GB vortex, some ridata and and a gskill 128 GB, a Intel X25 80 GB and a Intel 510. The latter hasn't been driven through the mill too much yet.

[[XC] Oj101]

Ok, I seem to be a very heavy user. I sold the small drives that I had in RAID0 so I'm back to AHCI mode and monitoring the SMART readings over the last few days I seem to be writing between about 10 and 85GB per day. I do think Opera could be a fair contributor to this, but on a 384kbps line I can't do more than about 3GB per day downloading 24/7. That being said, tabs hog memory and lately I can go through 500 tabs in an evening without thinking twice. The night before last I counted 247 tabs open simultaneously, this excludes dozens of tabs which had been opened and closed prior.



I have 4GB of RAM and Opera was taking 3GB for itself - that excluded Windows 7 which takes about 800MB, six threads of WCG which take about 500MB, GPUGrid at (I think) 100MB, Need For Speed at 1.4GB, FireFox at about 330MB... We're at 6.2GB total, or 2.2GB into the pagefile assuming every bit of RAM was used.

Other than that, I think World Community Grid is also fairly heavy. The discrepancy between running time and CPU time used to be about 45 minutes by the end of a thirteen hour work unit, but that figure is down to under four minutes since swapping to an SSD. To me that seems to suggest that WCG is very heavy with 4KB writes.

Since posting this topic six weeks ago, I've added 1.7TB of writes to the SSD.

[uOpt]

Did you try any of the suggestions?

You also mix up the concepts of virtual and physical memory, your memory calculations won't do.