dslreports logo
 
    All Forums Hot Topics Gallery
spc
uniqs
17

DKS
Damn Kidney Stones

join:2001-03-22
Owen Sound, ON

DKS to Freddy

to Freddy

Re: [WIN7] Recommended SSD System Tweaks?

said by Freddy:

trparky,

Here is the link to information posted over at the Crucial website concerning SSD tweaks and tips:

»forum.crucial.com/t5/Sol ··· d-p/4900

Freddy

Drive alignment is irrelevant for SSD's.
dave
Premium Member
join:2000-05-04
not in ohio

dave

Premium Member

said by DKS:

Drive alignment is irrelevant for SSD's.

Not so. SSDs read/write data based on their notion of an internal block size, and the OS read/writes data based on its notion of a cluster size. You don't want OS clusters misaligned with respect to SSD blocks, since otherwise one OS write can require modifications to two SSD blocks: performance goes way down.

But Windows 7 handles that correctly. Previous versions got it consistently wrong (because they started the first partition something like 63 sectors in from the beginning of the disk).

trparky
Premium Member
join:2000-05-24
Cleveland, OH

trparky

Premium Member

So do I even need a page file for a system that has 8 or 12 GBs of RAM?

Kilroy
MVM
join:2002-11-21
Saint Paul, MN

Kilroy

MVM

said by trparky:

So do I even need a page file for a system that has 8 or 12 GBs of RAM?

Yes, but move it to a mechanical drive, not the SSD. You most likely will not use the page file, but running without a page file can also cause issues. An alternative would be to have a small 1024 MB page file on the SSD. I recommend against using hibernation also as it will use the same size on the SSD as you have in RAM.

Vchat20
Landing is the REAL challenge
Premium Member
join:2003-09-16
Columbus, OH

Vchat20

Premium Member

said by Kilroy:

said by trparky:

So do I even need a page file for a system that has 8 or 12 GBs of RAM?

An alternative would be to have a small 1024 MB page file on the SSD. I recommend against using hibernation also as it will use the same size on the SSD as you have in RAM.

These are the only changes I would really recommend due to space concerns and what size SSD you have/plan to get. Reduce PF size to something reasonable. More than likely you will never need to use it or if you do, it will be minimal. Hibernate, as noted, will use close to or the same amount of RAM you have in the system (I have always assumed and read it was equal to the total RAM but have noticed on a few 7 systems that it is a bit less for some unknown reason. One in particular has 2GB but the hiberfile size is aroun ~1.7GB). -IF- you use hibernate, move it. If not, just disable hibernate altogether and the file will go away with it. In fact it would be my personal opinion to avoid hibernation on this machine anyways given how much RAM you have. It would likely take much longer to hibernate/resume even if the hiberfile is on the SSD, than to simply do a cold boot to begin with.

trparky
Premium Member
join:2000-05-24
Cleveland, OH

trparky

Premium Member

Should SuperFetch be enabled or disabled?

DKS
Damn Kidney Stones

join:2001-03-22
Owen Sound, ON

DKS to dave

to dave
said by dave:

said by DKS:

Drive alignment is irrelevant for SSD's.

Not so. SSDs read/write data based on their notion of an internal block size, and the OS read/writes data based on its notion of a cluster size. You don't want OS clusters misaligned with respect to SSD blocks, since otherwise one OS write can require modifications to two SSD blocks: performance goes way down.

But Windows 7 handles that correctly. Previous versions got it consistently wrong (because they started the first partition something like 63 sectors in from the beginning of the disk).

Fred Langa at Windows Secrets did comprehensive testing this week (unfortunately it's behind a paywall or I would link it) and showed that drive alignment makes zero difference in an SSD.
quote:
The throughput times of the misaligned-versus-aligned SSD produced the following results:
Random-access read
Before alignment: 149.2 megabytes per second (MBs)
After alignment: 150.0 MBs
Sequential read
Before: 278 MBs
After: 277 MBs
Windows Experience Index
Before and after: 7.9 (The scale goes only to 8.) Bootup and shut down
No perceptible change.
And so on

In every area I looked at, I could detect no meaningful difference in the SSD's aligned and misaligned performance.
These results make sense if alignment-performance issues are mostly related to drive mechanics. When a traditional platter drive is misaligned, its heads might have to make additional movements to completely read or write data to each sector. With a single file using hundreds or even thousands of sectors, those extra movements could add up to a significant amount of time.

SSDs, on the other hand, have no platters or heads. Whether the sectors are aligned or not, accessing different memory addresses in an SSD's RAM takes almost no time at all. So it's not surprising that aligning the sectors on my SSD showed no significant performance improvement.

Drive alignment might have better results with other configurations heavy, constant database lookups; conventional, mechanical drives; or RAID systems. Your results might also be different if your drive is formatted with tools that handle alignment differently or not at all.


JohnInSJ
Premium Member
join:2003-09-22
Aptos, CA

JohnInSJ

Premium Member

Is it just me or did Fred test the wrong thing.

If not aligned, WRITEs are a problem
- a write can cause two SSD blocks to need to be updated, which is an unnecessary update to a flash memory block. Life-1 for no reason
- a write update to two blocks is going to take more time than a write to one block

So, testing read speeds is going to show either of these things how, exactly?

DKS
Damn Kidney Stones

join:2001-03-22
Owen Sound, ON

DKS

said by JohnInSJ:

Is it just me or did Fred test the wrong thing.

If not aligned, WRITEs are a problem
- a write can cause two SSD blocks to need to be updated, which is an unnecessary update to a flash memory block. Life-1 for no reason
- a write update to two blocks is going to take more time than a write to one block

So, testing read speeds is going to show either of these things how, exactly?

You might find this thread helpful in response to Fred's comments:

»windowssecrets.com/forum ··· e-drives

Frankly, it seems to be, for most people, searching after the Holy Grail. And except for the very technical or extremely anal, not to amount to a hill of beans.

JohnInSJ
Premium Member
join:2003-09-22
Aptos, CA

JohnInSJ

Premium Member

The linked comments say exactly what I said, with more words

Write is the issue, not read. Slight but possibly insignificant in modern SSD increase in wear. Two read/writes vs 1... two REALLY FAST R/Ws, but still, double the time. For Writes. Sometimes. So Fred tested the wrong thing (read speed)

Vchat20
Landing is the REAL challenge
Premium Member
join:2003-09-16
Columbus, OH

1 recommendation

Vchat20 to trparky

Premium Member

to trparky
said by trparky:

So do I even need a page file for a system that has 8 or 12 GBs of RAM?

7 will pretty much handle this stuff automatically such as disabling automatic defrag on the disk, disabling superfetch, and other system services unnecessary on an ssd.

Just do a fresh install on the drive without going outside of 7's guided install process and do the prior mentioned pagefile and hibernation tweaks and you'll be fine. No need to overthink it.
dave
Premium Member
join:2000-05-04
not in ohio

1 edit

dave to DKS

Premium Member

to DKS
It so happens that I often chat to a senior guy where I work whose job in part involves figuring out the impact of various disk technologies on our product line. And he tells me it does make a difference. Unfortunately his reports aren't public.

Misalignment is mostly a write problem, and more so on a not-new drive. (You can't overwrite an SSD block; the controller needs to allocate an erased block, and schedule the old one for erasure.)

FWIW, not all SSDs are equal. Different controllers choose to optimize for (note rare example of correct usage of the word) different scenarios.

DKS
Damn Kidney Stones

join:2001-03-22
Owen Sound, ON

DKS to Vchat20

to Vchat20
said by Vchat20:

said by trparky:

So do I even need a page file for a system that has 8 or 12 GBs of RAM?

7 will pretty much handle this stuff automatically such as disabling automatic defrag on the disk, disabling superfetch, and other system services unnecessary on an ssd.

Just do a fresh install on the drive without going outside of 7's guided install process and do the prior mentioned pagefile and hibernation tweaks and you'll be fine. No need to overthink it.

+1
DKS

DKS to dave

to dave
said by dave:

It so happens that I often chat to a senior guy where I work whose job in part involves figuring out the impact of various disk technologies on our product line. And he tells me it does make a difference. Unfortunately his reports aren't public.

Misalignment is mostly a write problem, and more so on a not-new drive. (You can't overwrite an SSD block; the controller needs to allocate an erased block, and schedule the old one for erasure.)

FWIW, not all SSDs are equal. Different controllers choose to optimize for (note rare example of correct usage of the word) different scenarios.

In that situation, I can understand the requirement. But for the 99% of consumers, not so much. Let Windows 7 do its thing and the rest of us should be fine. After several drive failures (OCZ) I have been using Intel SSDs with no problems.

shearer
Northern Lights
Premium Member
join:2002-06-18
Asia

shearer to Vchat20

Premium Member

to Vchat20
said by Vchat20:

7 will pretty much handle this stuff automatically such as disabling automatic defrag on the disk, disabling superfetch, and other system services unnecessary on an ssd.

OK, but what if I'm restoring an image of Win7 (originating from mechanical HDD) to SSD? What other tweaks I need to make beside the following:

- drive alignment (if I want that extra non-perceptible boost in performance)
- disable superfetch, defrag, indexing
- move page file away
- TRIM (does win7 automatically enable it if it detects itself sitting on a SSD?)

BronsCon
join:2003-10-24
Fairfield, CA

BronsCon to trparky

Member

to trparky
said by trparky:

So do I even need a page file for a system that has 8 or 12 GBs of RAM?

Depends what you're doing with it. I run 16GB with no pagefile, but I'm gonna have to turn that puppy back on when I start running my VMs again. If you're just surfing the web and checking your email, you're probably good without a pagefile even at 4GB.

Turn off hibernation, set the pagefile initial size to 0, max size to whatever (usually equal to the amount of RAM you have is a good policy, double your RAM if you have 4GB or less) and Windows won't page out unless it absolutely needs to. Disabling the pagefile altogether means a crash if you fill your RAM, setting it to a 0 initial size means it won't be used unless your RAM is literally full; giving it a nonzero initial size means Windows will page out enough data to fill that initial size as soon as there's enough disk idle time for it to do so without impacting performance. It does this and marks the paged out portion of RAM as disposable, so it knows it can just free that RAM without having to page it our first (since it's already written to disk), but if you never actually fill your RAM, you're using erase-write cycles on your SSD needlessly.
dave
Premium Member
join:2000-05-04
not in ohio

dave to DKS

Premium Member

to DKS
said by DKS:

Let Windows 7 do its thing and the rest of us should be fine.

There's no alignment issue with Windows 7 - it knows how to do the alignment.

Previous versions of Windows did not; so those are the systems on which you need to fix the alignment (if you care). That includes the update from not-Win7 to Win7 case, of course.

digitalfutur
Sees More Than Shown
Premium Member
join:2000-07-15
GTA

digitalfutur to DKS

Premium Member

to DKS
Agreed. Even defragging sata drives is required only rarely (once or twice every 6 months).

Super fast CPUs and HDDs render the performance penalty of fragmentation as close to 0 as a mechanical device can be.

Sometimes ya just gotta let go...

MrWhsprs
Premium Member
join:2000-04-22
Round Lake, IL

MrWhsprs to Vchat20

Premium Member

to Vchat20
said by Vchat20:

-IF- you use hibernate, move it.

Just a quick note that the hibernate file can't be moved. It's my understanding that the hibernate file will always be (and can only be) in the root of the system partition, where ntldr is located.

trparky
Premium Member
join:2000-05-24
Cleveland, OH
·AT&T U-Verse

trparky

Premium Member

Click for full size
High Interrupts
And the damn monster is back. ARG!!!!

DarkLogix
Texan and Proud
Premium Member
join:2008-10-23
Baytown, TX

DarkLogix to BronsCon

Premium Member

to BronsCon
said by BronsCon:

said by trparky:

So do I even need a page file for a system that has 8 or 12 GBs of RAM?

Depends what you're doing with it. I run 16GB with no pagefile, but I'm gonna have to turn that puppy back on when I start running my VMs again. If you're just surfing the web and checking your email, you're probably good without a pagefile even at 4GB.

Turn off hibernation, set the pagefile initial size to 0, max size to whatever (usually equal to the amount of RAM you have is a good policy, double your RAM if you have 4GB or less) and Windows won't page out unless it absolutely needs to. Disabling the pagefile altogether means a crash if you fill your RAM, setting it to a 0 initial size means it won't be used unless your RAM is literally full; giving it a nonzero initial size means Windows will page out enough data to fill that initial size as soon as there's enough disk idle time for it to do so without impacting performance. It does this and marks the paged out portion of RAM as disposable, so it knows it can just free that RAM without having to page it our first (since it's already written to disk), but if you never actually fill your RAM, you're using erase-write cycles on your SSD needlessly.

Actually ideal for the pagefile is set min and max to the same size so it doesn't re-size it
DarkLogix

DarkLogix to dave

Premium Member

to dave
said by dave:

said by DKS:

Let Windows 7 do its thing and the rest of us should be fine.

There's no alignment issue with Windows 7 - it knows how to do the alignment.

Previous versions of Windows did not; so those are the systems on which you need to fix the alignment (if you care). That includes the update from not-Win7 to Win7 case, of course.

If in doubt check the alignment, its a few command line commands and you can see if its aligned.

if its not then fix
this applies to all SSD's and all 4k block drives (often called advanced format, as the 4k block size was needed to make drives over 2TB)

BronsCon
join:2003-10-24
Fairfield, CA

BronsCon to DarkLogix

Member

to DarkLogix
You're thinking spinning disk, where there are performance implications to fragmentation. Taking that out of the equation changes the game; the ideal pagefile is the smallest one you can get away with at any given moment.

DarkLogix
Texan and Proud
Premium Member
join:2008-10-23
Baytown, TX

DarkLogix

Premium Member

said by BronsCon:

You're thinking spinning disk, where there are performance implications to fragmentation. Taking that out of the equation changes the game; the ideal pagefile is the smallest one you can get away with at any given moment.

Re-sizing is extra IO, though if you have enough ram then set a small page file, I have 24GB of ram so I set a tiny pagefile with same min/max size so its there but doesn't do anything.

If it has to change the file size that is more IO than if it just had to update the data of the file, so its still a good idea for a non-changing size. (though if you have enough memory to not need a pagefile then it's going to sit there empty and not have an effect anyway.)

and short of 2 programs I use nothing has come close to filing my ram and even those 2 haven't fully filled it.
dave
Premium Member
join:2000-05-04
not in ohio

dave to trparky

Premium Member

to trparky
Your screenshot shows count, not rate. A high count is not necessarily a problem.

trparky
Premium Member
join:2000-05-24
Cleveland, OH

trparky

Premium Member

dave See Profile, what do you want to see? What graphs, data tables, etc. do you want to see?
trparky

trparky

Premium Member

Click for full size
High Interrupts v2
Ok, here's a new graph.
trparky

trparky

Premium Member

Click for full size
DPC Graph
And now the DPCs...