speeding up vista with Core

[Bansh33nl]

1 OCZ Core V2 30GB

Windows Vista 64bit ultimate optimized with vista manager full
Asus Maximus Formula
E8500 @ 4.3 Ghz

All the steps in the beginning of this thread were used.

Without Write caching:

http://www.plaatjesupload.nl/bekijk/2008/10/30/1225372429-460.jpg

With Write caching:

http://www.plaatjesupload.nl/bekijk/2008/10/30/1225372475-230.jpg

On my pc there is definetly a difference with write caching

Sorry for the big images. Dropped back to links...please 800x600max

By the way the ssd was formatted with the standard allocation size.

[Tony]

1 OCZ Core V2 30GB

Windows Vista 64bit ultimate optimized with vista manager full
Asus Maximus Formula
E8500 @ 4.3 Ghz

All the steps in the beginning of this thread were used.

Without Write caching:

http://www.plaatjesupload.nl/bekijk/2008/10/30/1225372429-460.jpg

With Write caching:

http://www.plaatjesupload.nl/bekijk/2008/10/30/1225372475-230.jpg

On my pc there is definetly a difference with write caching

Sorry for the big images. Dropped back to links...please 800x600max

By the way the ssd was formatted with the standard allocation size.

forget benches...does Vista with the tweaks feel faster to use?

[Bansh33nl]

Response is a bit faster thats all. Maybe the startup is 1 sec faster dunno. At least the ssd is faster than my old raptor 74gig. Can´t really veel alot of difference compared to my velociraptor 150gig. Will buy a second ssd next month.

[D111]

Download procmon from here,

http://technet.microsoft.com/en-us/s.../bb896645.aspx


Run it, see what is holding up your system: CPU, Disk writes, etc.

Come back and tell us where the issues are.

[Converterduck]

Hi brothers and sisters in SSD performance pain!

I have made similar painful experiences, but I think I have found a good solution to get optimal performance from SSD's with Windows XP or Vista.

As we all know are MLC SSD having their limitation when it comes to write performance with small random writes. While the read/write performance figures look impressive for most SSD's when it comes to IOPS (Input Output Operations per Second), the SSD's really suck. I purchased the product MFT (Managed Flash Technology Product) from Easy Computing Company. ( http://www.easyco.com/ ). You can download a 30 day test license and try it out yourself and see how it works for you. I did and then I just HAD TO PURCHASE IT as I would not want to live without it anymore.

I had great success with this software (Version 0.981 at time of writing this) because it significantly accelerates the random write performance and it also increases the life spawn of a Flash SSD.

The MFT product supports Linux and Windows operating systems. I use it to accelerate my 120GB OCZ Core Series V2 SATA II 2.5" SSD with Windows Vista SP1. It works perfectly and I will explain below I use it with my HP Tablet PC TC4400.

The MFT product that I purchased is MFT-SD128 for $110, money that was well invested. The Windows product does not support boot partitions. That posed no real problem for me since I always partition my laptop or desktop hard drives in a operating system partition and a data partition. In my case I use a 45GB OS partition and a 75GB data partition. I use MFT only on the data partition.

To increase the read performance on the Vista boot partition, I changed the cluster size from the default size to 64k. Since you cannot install Vista on a preformatted NTFS partition formatted with 64k clusters, you have to change the cluster size with a disk utility after installing Vista. I used Acronis Disk Director Suite 10.0 to change the cluster size and it worked without any problems.

I tried Paragon Partition Manager 8.5 but it did not work. Vista did not boot after changing to cluster size of 64k when using Paragon. I found in various notice boards evidence that Paragon has a problem converting correctly and Acronis works. A finding that I can confirm.

The OS/Program partition should give the best possible read performance and the data partition should give a balanced and fast read AND write performance, especially for random writes. Microsoft Outlook PST files are especially sensitive random I/O performance. Outlook is now blazing fast since the OST and PST files are placed onto the MFT powered data partition.

So since the boot partition cannot be used with MFT, you have to reduce the number of random write operations to the OS partition as much as possible. So I moved the TEMP / TMP for the users and the System to the data partition. Same for Firefox and Internet explorer caches. Same for the SWAP file. This has significantly improved the overall system performance.

I also implemented all the other good tuning tips that I found in another tread on this forum:
http://www.ocztechnologyforum.com/fo...ad.php?t=43525
such as disabling Vista indexing and search, disabling prefetch, disabling 8.3 filenames, etc.

Once I was satisfied, I wanted to get even better performance and I moved all temp files and internet chaches to a RAM Disk. There are several products on the market , but I made the very best experience with RamDisk Plus 9 from Superspeed ( http://www.superspeed.com/desktop/ramdisk.php ). This software costs $49.95.

I love this RamDisk Plus product because it allows me to make use of the 4GB RAM that I have installed in my notebook. A 32bit operating system like Windows XP or Vista cannot address the full 4GB of RAM, depending on hardware it can only address about 3.4GB, so you waste 600MB of RAM that just sits dead in your system. RamDisk Plus 9.0.3.0 can use the RAM above 3.4GB for RAM Disks. So creating a 600MB RAM Disk for your TEMP files and internet caches does not waste any of your valuable RAM.

In my setup, I have "invested" another 400MB system RAM to create a 1GB RAM disk (consisting of 600MB that could not be addressed otherwise + 400 MB of addressable RAM). I have now placed a 600MB SWAP file onto that RAM Disk. Here you can read about religious wars about how crazy it is to use precious system RAM for SWAP files. I know about all pro's and con's of doing this, but the setup that I described here works perfectly for me. Since I use my Tablet PC mainly for business applications, I never ran out of virtual memory with that setup and I do not have any swap file on a SSD partition that can wear and tear the MLC flash cells. :wpleased:

Benchmarks:
The MFT product comes with a good and comprehensive benchmark that is built into the software. It tests the read and write performance with different block sizes between 512Bytes and 4MBytes with 1, 10, and 40 treads. If you install the demo license you can use it to benchmark your drive. Since I have a 45GB C: partition for the OS without MFT and a 75GB D: partition WITH MFT, I couldmake comparisons between a standard 64k cluster partition and a MFT enabled partition.

The read speed of the standard 64k cluster partition is more than twice as fast as the read speed on the MFT partition, The write speed on the MFT partition is more than 4 times as fast as the write speed on the standard partitions. The number of possible IOPS on the MFT partition was more than three times the number of IOPS on the standard partition. The exact factor varies by block size and number of treads.

See copy and paste of my benchmark results in attached Word Document.

So I hope that this post will help you to take the maximum out of your SSD Flash drive. It would be great if OCZ would bundle the Managed Flash Technology (MFT) software with their products or offer it as an option together with the drive.

I am looking forward to your feedback and your experiences.

[geosoft]

Hi Converterduck,


In your benchmark data, random write speed of MFT volume is much slower than laptop hard disk.

Your MFT volume 4KB random write - 28 IOPS, 111~115KB/sec

My Hitachi laptop hard disk 4KB random write - 217 IOPS, 858~889KB/sec

In my setup, random write speed of MFT volume is much faster than yours.

4KB random write - 12,881 IOPS, 50MB/sec

X 100 faster than yours!!!


http://www.ocztechnologyforum.com/fo...ad.php?t=45030


How much kilo-bytes did you set as erase block size when mounting MFT volume?

Easyco recommend setting erase block size of OCZ Core to 16MB(16,384KB).

I had set erase block size of my Core V1 64GB to 8MB(8,192KB).

Unless you had set erase block size other than 8MB or 16MB, try that.

[Converterduck]

As I mentioned in my previous post, you can also use the built-in benchmark function of the MFT software to benchmark other hard drives. I installed a demo license of MFT on my 3 year old custom built server that is based on a ASUS motherboard A8N-SLI Premium with a AMD 64 X2 4400+ CPU. I benchmarked four different conventional hard drives:
PATA Western Digital WD5000AKB
SATA RAID 0+1 Array with 3 x Samsung HD300LJ
PATA Samsung SP2514N
SATA Western Digital WD10EACS

I attached the benchmark results in the enclosed word document so you can compare it with the SSD benchmark that i did for my OCZ Core Series V2 SATA II 2.5" SSD with Windows Vista SP1. Hope you find this informative.

I would be interested in your benchmark results that you get with MFT with your SSD installation .

[Converterduck]

Hi Geosoft
Maybe by tablet pc is a bit slow as it is 2 years old.
I followed Easyco's recommendation and set an erase block size of OCZ Core to 16MB(16,384KB). So maybe I should try with 8MB and try again?

[Converterduck]

I attached a screen copy (JPG file) of page 16 from MFT installation manual. As you can see for "KB Write Block Size" for OCZ core V1 and V2, for which it recommends 16384. I have now tried with reducing it to 8192 and WOW!!

The difference is SIGNIFICANT compared to my previous results. MFT rocks with a KB Write Block Size of 8192.

with 16384: 4k Random Write 111-115kB/sec @ 28 IOPS
with 8192: 4k Random Write 15.3 - 26.3 MB/sec @ 6500-6700 IOPS

And its interesting that the 4K Random Read performance has also significantly increased from 123 KB/sec @30 IOPS to 8.1 MB/sec @ 2067 IOPS


Thanks GEOSOFT for the tip with KB Write Block Size!

See attached screen shot of MFT Benchmark.

[trismegistosh]

Caching random writes to RAM before committing them to an SSD improves performance.

I believe the main reason for the performance improvement over non-cached writes lies in the fact that SSDs perform significantly worse in mixed random read/write scenarios, compared to doing only random reads and only random writes. Up to 75% performance degradation has been mentioned for certain SSDs in mixed random read/write scenarios, compared to only doing random reads and then only random writes.

E.g. doing 100 MB of random reads, followed by 100 MB of random writes will go much faster than doing 200 MB of random reading and writing interleaved (still portioned 50/50).

Caching writes to RAM lessens the need for an SSD to interleave random reads and random writes, thereby avoiding the performance degradation that follows from continuously mixing random reads and random writes.

That's why I use ram overlay with EWF to cache all writes to RAM then commit them afterwards. Lazywrites or write behind disk caching may give same benefits but personally I will go the ewf route barring the memory limit.

[D111]

Just a short note to thank everyone for sharing their experiences and tips.

SSD technology is still at a point where big improvements in performance can be made in tweaking and tuning, and in mating hardware together properly....

Like to good old days of the IBM PC with 16k of DRAM onboard.

I almost dread the day, maybe 7 years hence, when there will be as little improvement possible as from tweaking DRAM...

Carry on and enjoy yourselves...

[therookie]

That's why I use ram overlay with EWF to cache all writes to RAM then commit them afterwards. Lazywrites or write behind disk caching may give same benefits but personally I will go the ewf route barring the memory limit.

EWF is very nice from a performance perspective. I have been using it for my USB Windows XP installs and it has resulted in fantastic performance improvements, eliminating OS stuttering and stalls among other things.

However, while many tweaks (including EWF) in these forums are excellent performance enhancements, they also carry significant risks and downsides. While some tweaks only eliminate functionality (e.g. disable system restore) others introduce the risk of losing significant amounts of data (e.g. EWF).

These risks and downsides are not a problem to enthusiasts (as they can manage them) but to the non-enthusiast users they are in many cases showstoppers (this can be seen in some of the responses to some of the proposed tweaks in these forums).

I believe that there are two different solutions to the stuttering and performance problems of the Core Series drives. One solution is the combination of the many tweaks and fixes presented in these forums. They give unparalleled performance increases but also do it at the expense of functionality and data safety.

The other solution (for the non-enthusiasts) is to fix the OS stuttering and stalls without sacrificing functionality or data safety (or at least to a significantly lesser extent). Naturally, that type of solution will not be able to parallel the performance enhancement of the first solution. However, the Core series drives are pretty darned fast in all other aspects so non-enthusiast users will likely be happy as long as they can avoid stuttering/stalling and long installation times.

I believe that the proper combination of write caching / lazywriting / write pipelining is the key to achieving the second solution. It looks like I might be getting access to one or two Core series drives in the near future. If that happens, I will try to replicate the reported stuttering problems (e.g. Outlook 2007 with large .ost and .pst files) and then try resolving them without resorting to EWF and ramdisk-style fixes.

[D111]

@Rookie

My recommendation is to lay out clearly the Risk / Reward table, and then gear it to the users specifically.

If a user is using it as a workhorse, that may be a low risk, but accept modest gain environment. That is kind of like my main machine --- geared for safety and reliability.

Users that are running desktop screamers that run games may be willing to take a higher risk.

In order to replicate the Outlook problems, create a dummy PST file that is at least 1gb big --- easiest way to do it is to sign up for a ton of news groups, and wait 24 hrs for the thing to fill with messages.

Then do some dummy records in address file.


Good luck.

[therookie]

I agree.

I was thinking in the lines of producing a tweak installer where the user can either make a general selection i.e. High performance increase (but with highest “cost”) vs moderate performance increase (with no or lower “cost”), as well as a manual mode where one can enable/disable each tweak individually and where the pros and cons of each tweak are stated.

"Unfortunately", I have some +6GB mail in Outlook so subscribing to newsgroups is the last thing I need ;-).

[geosoft]

Another replication of freezing and stuttering

http://www.ocztechnologyforum.com/fo...ad.php?t=44934


Repeatedly check the time for compressing and uncompressing a large folder that contains many files.

It's not regular.

the more freezing and stuttering --> the longer