Readyboost sucks, my way is better

[ShawnD1]

Many of us have found out the hard way that readyboost doesn't actually do anything to improve performance. Although readyboost tries to read from flash, Windows is still writing the swap file to the hard drive as well as the flash. If you're trying to access the hard drive while it's writing to the swap file, the computer drags ass and is virtually unusable. This is why readyboost doesn't work, but it can be fixed with a little bit of tweaking.

Step 1: force Windows to see your USB drive as fixed
Changing the "removable media bit" on a flash drive is done with a program called BootIt. Download the program from this page (the link is on the bottom left). Run the program, select the drive letter of your USB drive, hit the "flip removable bit" button. Here's a picture:
http://img189.imageshack.us/img189/9895/bootit.jpg

Now the USB drive should appear as a fixed drive in windows explorer. In this picture, G is a flash drive:
http://img188.imageshack.us/img188/2176/localdrive.jpg

Step 2: put the swap file on the USB drive
take the swap file off every drive except the USB drive
Here is a picture:
http://img46.imageshack.us/img46/6669/pagefile.jpg

Step 3: restart computer
restart the computer...

Step 4: check if it worked
Open performance monitor (vista only) and check if your USB drive has any activity on a file called "pagefile.sys"; here is a picture:
http://img132.imageshack.us/img132/7651/perfmon.jpg


And now the computer will run faster. My file server is quite a bit faster after doing this because all of its drives are cheap shitty drives; the ones that are $100 for 1000gb. If your system is like mine and it contains literally the cheapest hard drives you can buy, putting the swap file on a USB drive makes a world of difference.

 

[zerocool84]

Or you can put in another GB of RAM for ~$10 and not have to worry about it.

 

[ShawnD1]

Originally posted by: zerocool84
Or you can put in another GB of RAM for ~$10 and not have to worry about it.

The swap file is critically important to how the OS works, and it can't be replaced with more memory. Not only does the swap file pre-allocate address space, but data naturally goes back and forth between the ram and the swap based on whether or not a program is being used. Have you ever started a game, played it for a few minutes, minimized it for some reason, and came back a few minutes later to find that the program was incredibly slow? What happened was the OS noticed you weren't using that program, so it moved game data out of the memory and into the swap. The slower your swap is, the longer it takes for the data to get back into memory.

You can check how heavily your swap file is being used by going into Task Manager. The "commit charge" of a program is the total amount of space that program has and "working set" is the amount of real memory it has. Some programs are mostly memory, and some are mostly swap. For example, Folding @ Home is using 10mb working set but its commit charge is just over 80mb. The amount of swap being used far exceeds the amount of ram being used even though I have 2gb of ram that is unused.

The computer I'm on right now has a total of 3.2gb usable memory, task manager says 1.2gb of that is being used, and performance monitor says the file with the highest rate of data being written to it is C:\pagefile.sys

 

[bsobel]

The swap file is critically important to how the OS works, and it can't be replaced with more memory. Not only does the swap file pre-allocate address space, but data naturally goes back and forth between the ram and the swap based on whether or not a program is being used. Have you ever started a game, played it for a few minutes, minimized it for some reason, and came back a few minutes later to find that the program was incredibly slow? What happened was the OS noticed you weren't using that program, so it moved game data out of the memory and into the swap. The slower your swap is, the longer it takes for the data to get back into memory.

Actually that is no longer true Vista onward, you can live with out a swap file and the system will allocate a temporary ones for the cases where XP would cause an error (such as null memory mapped files).

 

[spikespiegal]

I have too many questions about this, but to say I'm skeptical is an understatement.

First, why use Vista as a file server? Also, given that virtual memory shouldn't be an issue with background tasks such as network file services why would you see a performance increase in the first place? Maybe if I were running NT4.0 on 32meg of RAM (nevermind).......you do have more than 512meg in this box, don't you?

The Windows swap file has always been extremely sensitive to write latency, which is why a lot of SANs engineers still insist that the best place for the swap file is on a local drive. Simply put, I dare say the performance of many of the SANs I've worked on is a lot faster than any USB drive you have.

If your swap file is otherwise doing a lot of grinding, then there's another issue that needs fixing. Having a large drive filled up over 75% and fragmented to hell is one way to cause swap file thrashing. In that respect moving the swap file to a USB drive, or *any* other location for that matter will result in a performance increase. Your typical USB drive isn't a performance match, or even close, compared to 3-4 year old IDE/SATA drive that's reasonably clean, so there's some other reason for your performance increase.

 

[ShawnD1]

Originally posted by: spikespiegal
First, why use Vista as a file server?

Because it's the only suitable server OS I've found.

Also, given that virtual memory shouldn't be an issue with background tasks such as network file services why would you see a performance increase in the first place? Maybe if I were running NT4.0 on 32meg of RAM (nevermind).......you do have more than 512meg in this box, don't you?

It has 2gb of ram.

The Windows swap file has always been extremely sensitive to write latency, which is why a lot of SANs engineers still insist that the best place for the swap file is on a local drive. Simply put, I dare say the performance of many of the SANs I've worked on is a lot faster than any USB drive you have.

Actually if you test drive speed in Sandra you'll find that nand flash is much much faster than a standard hard drive. My 2gb JetFlash Transcend has a random access time of 255us, random read and write speed of 19mb/s. That same exact test done on a 1TB WD Caviar Green has an access time of 16ms (that's about 63x slower).

If your swap file is otherwise doing a lot of grinding, then there's another issue that needs fixing. Having a large drive filled up over 75% and fragmented to hell is one way to cause swap file thrashing. In that respect moving the swap file to a USB drive, or *any* other location for that matter will result in a performance increase. Your typical USB drive isn't a performance match, or even close, compared to 3-4 year old IDE/SATA drive that's reasonably clean, so there's some other reason for your performance increase.

Hard drives are not good for paging because the latency is too high. If it needs to waste 16ms getting to the swap file then waste another 16ms to get that file I told it to get, that's 32ms of seek time. If we assume the drive was trying to do a read operation at its full speed of 50mb/s (according to Sandra), 32ms of lag translates into 1.6mb of lost bandwidth, and that's just for a single write to the swap. If it's being written to a spare drive that is dedicated to only handling swap, the system has no slow down at all. The maximum swap file for Win32 is 4095mb; flash drives are a perfect fit.

 

[MadRat]

Flash memory performance will steadily degrade in the first 3 months of using it for ReadyBoost. Probably faster using it as a "swap" space. Tell us how good this method is in a few months.

 

[bsobel]

Actually if you test drive speed in Sandra you'll find that nand flash is much much faster than a standard hard drive. My 2gb JetFlash Transcend has a random access time of 255us, random read and write speed of 19mb/s. That same exact test done on a 1TB WD Caviar Green has an access time of 16ms (that's about 63x slower).

Seek times are most definately faster, sustained read/write times are not and are usally slower (much) slower. SSD's address this with their chipset, but plain flash drives do not.

 

[ShawnD1]

Originally posted by: MadRat
Flash memory performance will steadily degrade in the first 3 months of using it for ReadyBoost. Probably faster using it as a "swap" space. Tell us how good this method is in a few months.

That 2gb stick operating at 19mb/s has been used for readyboost since June 2007. I bought it the same day I bought Vista

 

[Jeff7181]

The ONLY reason I can see using flash memory specifically for Readyboost is if it's impossible to add more RAM to your computer. But if your computer has 2 GB and it's not fast enough, put 4 in... if that's not enough, put 8 in... if that's not enough, put 16 in. At $10 per GB, I'd have to say you'd be stupid to choose flash memory over system RAM given the choice.

 

[ShawnD1]

Originally posted by: Jeff7181
The ONLY reason I can see using flash memory specifically for Readyboost is if it's impossible to add more RAM to your computer. But if your computer has 2 GB and it's not fast enough, put 4 in... if that's not enough, put 8 in... if that's not enough, put 16 in. At $10 per GB, I'd have to say you'd be stupid to choose flash memory over system RAM given the choice.

You're forgetting the $200 cost of buying an OEM copy of a 64-bit OS

edit:
Actually, since this computer is already built, OEM is illegal. It would need a retail copy which is even more expensive.

 

[Jeff7181]

Originally posted by: ShawnD1

Originally posted by: Jeff7181
The ONLY reason I can see using flash memory specifically for Readyboost is if it's impossible to add more RAM to your computer. But if your computer has 2 GB and it's not fast enough, put 4 in... if that's not enough, put 8 in... if that's not enough, put 16 in. At $10 per GB, I'd have to say you'd be stupid to choose flash memory over system RAM given the choice.

You're forgetting the $200 cost of buying an OEM copy of a 64-bit OS

Because a 64-bit OS costs more than a 32-bit OS?

*EDIT* No, it wouldn't be illegal... it would be illegal to sell an OEM license without hardware... but you can buy a $5 case fan and an OEM license and it's legit... if it wasn't, Microsoft wouldn't let newegg do that.

**EDIT** And by the way, it's $100, not $200.

 

[ShawnD1]

Originally posted by: Jeff7181
**EDIT** And by the way, it's $100, not $200.

That is not a server OS. It lacks remote desktop.
The cost of upgrading the ram would then be $140 for the OEM Vista then $40 for 4gb of ram since the motherboard doesn't have any free ram slots.

Total cost to upgrade ram: $180
Total cost to "fix" readyboost: free

 

[Jeff7181]

Originally posted by: ShawnD1

Originally posted by: Jeff7181
**EDIT** And by the way, it's $100, not $200.

That is not a server OS. It lacks remote desktop.
The cost of upgrading the ram would then be $140 for the OEM Vista then $40 for 4gb of ram since the motherboard doesn't have any free ram slots.

Total cost to upgrade ram: $180
Total cost to "fix" readyboost: free

Who cares if it's not a server OS? Vista Business and Ultimate are not server OS's either. What does that have to do with anything? You're talking about alternatives to Readyboost. The absolute best alternative to Readyboost is more RAM, period.

And if you're proposing using a USB flash drive to host the swap file on a server... :roll: That may work for a hack job in your basement... but it won't fly for anything with even the smallest expectation of reliability. Your solution makes no sense... you're talking about server operating systems and consumer level USB flash drives... I can't even find the words to describe how ridiculous that is.

 

[ShawnD1]

It would help if you read the first post in the thread before you reply.

Originally posted by: ShawnD1
My file server

Do a google image search for what a "server" is. You'll notice that every single picture is that of a computer with no monitor, no keyboard, no mouse. Then you propose that such a headless system use an OS with no way to remotely control it. When called on your bullshit, you claim Vista cannot act as a server (even though it has been doing so for the past 2 years) and offer no suggestion as to what is a server OS.

You make yourself look even more silly by stating that flash is untrustworthy despite the fact that flash paging was touted as a major improvement in Vista and we have countless articles on Anandtech talking about how reliable SSD (flash) hard drives are because they are immune to shock damage, bearing failure, and all of the other fun stuff that has been plaguing hard drives since the dawn of computing.

Let's ignore all of that. Let's assume you're right and I can use Vista Home Basic on a headless computer. Your suggestion still costs a minimum of $140 to upgrade the memory to 4gb. The whole point of this thread is to improve your system performance for free. If I wanted to tell everyone how to make their slow system faster with no regard to cost, I would say "buy a new computer" and that would be the end of it.

 

[Jeff7181]

Lets not assume, agreed? I never said Vista can't act as a server, I said it's not a server OS. That's a fact. Windows Server is a server OS. The fact that Vista Business and Vista Ultimate can host RDP sessions doesn't make them a server OS.

I also never said Vista Home Basic can or should be used on a headless computer, I don't know where you came up with that.

Regardless... USB flash drives are NOT reliable enough to host the swap file, I don't care what you say. There's a reason Windows doesn't support putting the swap file on removable storage. They're consumer grade devices and are not designed to be reliable enough to store system files... especially connected via USB. Readyboost takes this into account and the OS does not count on the drive or anything it has stored on it being available.

You said your server has 2 GB of memory. Think about it... when that 2 GB is in use, it either hits the hard drive or in your case, a flash drive. Both of which are many orders of magnitude slower than system RAM... which you could replace with 4 GB for $40 without changing your OS. It makes absolutely zero sense to decrease system reliability by moving a system file to an unreliable storage medium to gain a slight increase in performance when you could spend a little more money (flash drives are not free) to increase performance without decreasing the reliability of this "server."

*EDIT* For the record, I don't believe Microsoft actually differentiates between 32-bit and 64-bit licenses. You can download (or borrow) a Vista Ultimate 64-bit disc image, install the OS and the key that came with your 32-bit copy will work fine.

 

[Rottie]

Originally posted by: MadRat
Flash memory performance will steadily degrade in the first 3 months of using it for ReadyBoost. Probably faster using it as a "swap" space. Tell us how good this method is in a few months.

maybe that could explain why my sandis cruzer micro died all of sudden after 5 months.

 

[Nothinman]

I never said Vista can't act as a server, I said it's not a server OS. That's a fact. Windows Server is a server OS. The fact that Vista Business and Vista Ultimate can host RDP sessions doesn't make them a server OS.

I've always hated those labels. The only real difference between MS' client OSes and Server OSes are the default settings and some licensing crap.

There's a reason Windows doesn't support putting the swap file on removable storage.

Actually I'd guess it has more to do with the fact that they're removable than with their reliability. If you pull a drive with a pagefile on it and Windows doesn't have that data stored anywhere else it has no choice but to kill the process asking for that data since it's gone.

 

[VinDSL]

Originally posted by: ShawnD1
Many of us have found out the hard way that readyboost doesn't actually do anything to improve performance...

If your system is like mine and it contains literally the cheapest hard drives you can buy, putting the swap file on a USB drive makes a world of difference.

LoL! You know what?!?!?

I've been lurking in this thread since you posted it - and I was inclined to call bullsh!t on it, but, now that I've thought about it for a while...

The biggest reason I use ReadyBoost is because it cuts the Vista HD chattering (on my laptop) down to zero - once the flash drive is fully populated. That is, when I boot my lappy into Vista, it takes about 5 minutes to load the files into my ReadyBoost flash drive, then my HD basically goes dormant.

To put it another way, I could care less if ReadyBoost gives my lappy a boost in performance. I use ReadyBoost to save my sanity! The HD chatter was driving ME crazy...

However, now that I *think* about it - I'm running 4GB RAM and 4GB ReadyBoost.

Really, There is NO reason to run 4GB ReadyBoost - 1 or 2 GB is plenty (for my stated purpose). The only reason I'm running 4GB ReadyBoost is because - what am I going to do with the other drive space, if I don't use it for ReadyBoost???

Well... I think you just came up with a good answer!

I'm going to give your idea a shot. I'm going to tell Vista to only use 2GB for ReadyBoost, and I'm going to use the remaining 2GB for my swap file.

What say you - think that will work???

 

[Joeygates]

Hi Shawn,

Your posts got progressively ruder and ruder as time went on. You veil it with a mask of civility but rude is still rude.

Your post title should bring this with no surprise to me.

I remember when these forums used to go without this sort of crap.

 

[dguy6789]

OP: This is faster
Skeptics: I'm not so sure it's faster
OP: Well it is faster
Skeptics: Still not sure about it

Is there any way to measure the performance gains of this in an accurate manner rather than relying on it feeling faster or slower?

 

[GodlessAstronomer]

I'm simply amazed that someone would be running a server box and be concerned about the price of a 64-bit OS and a couple of sticks of memory. This is the kind of thing that defies belief.

 

[Goosemaster]

ShawnD1,


While computer terminology and actual industry products clash, there are almost always a guiding set of principles that govern the basics for certain topics. On these forums you will find a myriad of enthusiasts and professionals whose cultures do not mix. please understand that some of us have explicit definitions for devices such as servers that go far beyond capabilities and feature sets.

File servers are some of the most basic application servers that nearly any OS can run. In this day in age if your OS supports fileshares using CIFS/SMB, NFS, ftp, or other similar protocols, it can be a fileserver. In general you can choose any OS, setup a fileserver, and set and forget it, managing it locally or remotely when needed.

With regards to your comments on speeding up file servers by leveraging the perceived speed of flash memory, I feel that you need to do more research. In researching not only the role of a fileserver but the requirements necessary to optimize data transfer, you will undoubtedly learn more about the role of swap data on a fileserver with adequate ram. in addition, you will hopefully learn the hindrance that a multipurpose OS, such as Vista [consumer version of the Windows OS line with NT6 Kernel] can be to a busy fileserver. There is a wealth of information out there that demonstrate the benefits of benefits of specialization based on server roles and the drawbacks of using non-specialized equipment. As for flash memory, please make sure to learn how to differentiate between flash-based USB thumbdrives and SSD's, what type of controller confirguartion they have, and how they perform when it comes to data access and transfer performance. I would also lookup the drawbacks of current SSD's so that you can begin to map the progress of flash memory, from the era of single-controller devices forward.

Hopefully on your journey you will find out why you are facing such resistance in promoting Vista and flash drives for server usage.


You are definitely on the right track in focusing on the dramatic access time deltas that can be achieved by using flash memory but the situation is indeed more complicated that it seems on the service, and oddly enough, the solutions much more simplistic


as an example of a rebuttal to your current arguments I would offer up available solutions such as freeNAS which use only 32MB of storage space, minimal ram and unnecessary services. once you do some research it will be easier for you to see the pros and cons of increased specialization and how it compares to solutions using generalized components such as Vista. In addition, you will hopefully discover the man drawbacks of current and past thumbdrives and why they are not as suitable as they appear to be.

Keep your head up and research this topic some more. flash memory usage on servers is really cool and I definitely urge you to to try learning a little more on the topic before you continue arguing.

 

[BD2003]

I tried this a LONG time ago...it didnt work very well. First off, readyboost has actually much less to do with the swap file than most people think it does.

Most usb sticks, compared to a hard drive, have:

1) very fast random reads
2) slow sequential reads/writes
3) horrendously slow random writes
4) total lack of a DRAM cache (which immensely helps to buffer small writes)

So if you actually run out of memory, and need to essentially run off the flash stick as memory, your PC will start getting really glitchy and freeze all the time as the I/O queue just grows and grows due to the random writes. There might be some situations where its a bit faster, but overall, if you actually rely on the swap file a lot, rather than just doing this for the shit of it and saying "hey, it's faster", you'll prob want to stick with readyboost, or even better, REAL memory.

Here's why readyboost is better:

It preloads useful data to the USB stick (like superfetch) - the swap file is basically leftovers.
It'll allow the hard drive to read data if it's faster - as in, if it detects a long sequential read, it'll get it from the HDD and skip the stick.

So basically, the system can write to the faster writing hard drive, write to the USB stick at leisure, and then read back from the faster reading USB stick. Its the best of both worlds.

Unless you actually have some proof that your method works otherwise. But ask yourself this - if your method is so ideal, why has it taken several years for someone to come up with it, when there's thousands of people, even microsoft engineers working on performance, who havent publicly came out with it first?

And it's still no replacement for real memory.

 

[Nothinman]

Is there any way to measure the performance gains of this in an accurate manner rather than relying on it feeling faster or slower?

No, because the impact of the pagefile's performance is affected by too many other things. There's no way to benchmark everyone's individual workloads.

I'm simply amazed that someone would be running a server box and be concerned about the price of a 64-bit OS and a couple of sticks of memory. This is the kind of thing that defies belief.

Not really, if I was building something for myself at home I'd want to spend as little as possible.