Hard disk (mech) latency difference per partition based on disk placement usage.

[Goildzy]

From top left to the right you have a 7200rpm 1TB hdd. After that in the top right corner is a SSD that is not relevant in this test. At the bottom we have a 5400rpm 500 GB HDD.

 

 

Tested all like: Main and main, mid and mid, low and low. 

 

These are serious drops over almost half the r/w speed on the 7200rpm. 25% ish for the 5400rpm. 

 

Based on this the middle partition should work the fastest while using multiple combinations next to the same partition so a mid+main or mid+low should have insane drops on main and or low because they have to reach so far inside to grab a piece of the middle partition (big header movement).  Where as the the middle partition only need to grab a small part of any of the other partition to read out while multi partioning with the neighbor. Results are that the middle partition barely loses speed in combined use to a neighbor.

 

Main+low should have a same like reaction but a different theory. The slow part of the disk is only 25% and so the slow reading can be done so fast that the header moves so much more on the main partition and has plenty of speed left to fill that little inner circle of the disk and return before even or barely any loss in speeds (probably some) 

 

Use of main+mid+low shall result in full disk usage and ull see a main till low drop in speed based on the disk part outer to inner. So main will have high and low will have the lowest ofcourse at the loss of the full speed potential of the disk. This Happens when your disk is full for example and the performance hit is big. 

 

(shall test this and uploas results) 

 

7200 main+mid-

 

7200 mid-low-

 

7200Main+low-

 

7200 all-

 

Now 5400

 

5400 main+mid

 

5400 mid+low-

 

5400 main+low-

 

5400 all-

 

Seems I was correct. Im glad if this is useful for you.

 

 

 

[Goildzy]

I calculated the space based on disk percentage usage and trying to isolate the middle of the disk by grabbing 25% and a little less of the the disks percentage. These tests have been running individually so not 2 partitions at once at the same on a disk.

 

Use the HDD 's main part with the fastest speed for your games and programs and other things that require a big performance reading like newer games. The mid part is reserved to be the next installment for your heavy working programs this is made to isolate the latency (not if you use 2 programs on 2 different partitions ofcourse it will make the head move from one to another and this is the biggest latency in mech hdds). The storage part speaks for it self. This is the slowest part of your HDD. You do not want this part at work with your current programs AT ALL to avoid latency. However you still have a huge value for STORAGE SPACE that sits still. And once files will transferred you wont lose alot of speed on the storage part. (the middle of the harddisk is the slowest part). 

 

The mix up should be.

 

SSD=OS, PROGRAMS/SERVICES/DRIVERS. 

 

BEST HDD Main =TOUGHEST WORK.

BEST HDD Mid=TOUGHEST WORK AFTER MAIN

BEST HDD Storage (low) =Still files with no movement until needing to move. 

 

LESSER HDD Main=WEAKER TOUGHEST WORK

LESSER HDD Mid=" see 7200rpm mid" 

LESSER HDD Storage (low) =" see 7200rpm low" 

 

PRIMARY programs need to run isolated per HDD and Partition to avoid head movement. Therefor play music from a different HDD and not mixed up with you main game/programs/work HDD, or stuff like any media low work... Video (not browser related due of being on SSD) netflix, etc those kinds use a different HDD if you prefer to multi program like game and watch programs or listen to music... You get my point. 

 

Pograms like older slower programs or weaker games  that don't really need 7200rpm at highest latency are better to be put on your other HDDs main even if thats a 5400rpm. Now you can run and old and or weaker program/game/work without head movement disturbing your work and speed/lag/stutter/latency drops etc. 

 

Partitions are always ordered in terms of speed by the OS, henge why the OS always suggest to install on your first partition. 

 

Plz correct my Englisch cause I might make mistakes. Any questions? Feel free to ask. 

 

+*Will add mixed partition results to this*+   !DONE SEE MAIN POST! 

[mariushm]

You can use HDTune to test throughput and access times, the trial version will give you a result like this :

(this is a 2.5" laptop drive, old, the speeds are normal)

 

 

The yellow dots are the access times - on my picture you see it's on 10ms increments (on the right side of the chart) and on the left side of the chart you have the throughput increment - my picture has 25 MB/s steps and 10 ms steps, yours may change depending on window size and speeds.

 

 

 

Nothing new, but there's no need to create partitions.

Simply using a good defragmenting program to move all executables and DLL files at the start of the drive (if you have a single partition that occupies the whole drive) would be enough.

 

For example, O&O Defrag allows you to specify file extensions that you want the software to move right at the beginning of the drive, file extensions to move all the way to the end (ex video files, music files, stuff that doesn't change often and is usually streamed or copied rarely)  , and file extensions that should be in the middle of the drive (files that shrink or enlarge often, that are edited often, in the center you'd have lots of free area allowing files to expand without fragmenting)

 

In the past, people did "short stroking" - basically creating a smaller partition on a drive in order to force a database application or some application to only use a tiny area of the drive, which would guarantee the drive heads will always move just a very tiny amount between a few tracks, instead of going all the way across the whole surface of the platters.

 

This way, they guaranteed a minimum latency when data had to be retrieved ... for example they took a 10k rpm 150 GB WD Raptor drive and created a 5-10 GB for the OS and 40 GB partition for a database. The drive had 8ms seek time if the head was all the way to one side and it had to seek a track on the opposite side, but if the drive heads had to seek only from left to right of that 40 GB area, the latencies could be below 5ms

 

 

For comparison, here's a SSD on the same laptop (that tiny drop was me opening several tabs in browser as I write this):

 

 

 

[Goildzy]

39 minutes ago, mariushm said:

You can use HDTune to test throughput and access times, the trial version will give you a result like this :

(this is a 2.5" laptop drive, old, the speeds are normal)

 

 

The yellow dots are the access times - on my picture you see it's on 10ms increments. On the left you have the throughput.

 

Nothing new, but there's no need to create partitions.

Simply using a good defragmenting program to move all executables and DLL files at the start of the drive (if you have a single partition that occupies the whole drive) would be enough.

 

For example, O&O Defrag allows you to specify file extensions that you want the software to move right at the beginning of the drive, file extensions to move all the way to the end (ex video files, music files, stuff that doesn't change often and is usually streamed or copied rarely)  , and file extensions that should be in the middle of the drive (files that shrink or enlarge often, that are edited often, in the center you'd have lots of free area allowing files to expand without fragmenting)

 

In the past, people did "short stroking" - basically creating a smaller partition on a drive in order to force a database application or some application to only use a tiny area of the drive, which would guarantee the drive heads will always move just a very tiny amount between a few tracks, instead of going all the way across the whole surface of the platters.

 

This way, they guaranteed a minimum latency when data had to be retrieved ... for example they took a 10k rpm 150 GB WD Raptor drive and created a 5-10 GB for the OS and 40 GB partition for a database. The drive had 8ms seek time if the head was all the way to one side and it had to seek a track on the opposite side, but if the drive heads had to seek only from left to right of that 40 GB area, the latencies could be below 5ms

 

 

You have not read everything it really applies to the partitions. If you u have read the use of services/os/dll/etc you'd see it would be seperated form any of the following mechinacal hdds. My test does not lie i am 100% right you can re apply the tests yourself. Using 2 partitions aka a lil mp3 and another file on your other partition will instantly devide your usage. Defragment wont help you here anyway...  The test you do applies to full use of a hdd aka All PARTITIONS are at work together. An ssd does not apply to mechanical movements so um yes that is an irrelevant state you van create 1000000 partitions on an ssd and wont lose performance. You can see my test result to see you are totally wrong. Your test need to be a file benchmark atleast know how to bench... 

 

Please remove your false information. 

[Goildzy]

1 hour ago, mariushm said:

*snip*

You spread false information and wrong benchmarks. Or know what the OP talks about... There you go your own benchmark software tells me im right.

 

Main-

 

Mid-

 

Low-

 

An inner part will never be equal in speed to the outer and the slightest  movements is insane latency drop. Again im 100% right read my article plz. 

Edited October 26, 2019 by wkdpaul

[mariushm]

33 minutes ago, Goildzy said:

You have not read everything it really applies to the partitions. If you u have read the use of services/os/dll/etc you'd see it would be seperated form any of the following mechinacal hdds. My test does not lie i am 100% right you can re apply the tests yourself. Using 2 partitions aka a lil mp3 and another file on your other partition will instantly devide your usage. Defragment wont help you here anyway... 

Well, I didn't read, but now that you replied I read and I can tell you that you're wrong.

 

You will always get faster transfer speeds towards the interior of a drive, simply because the actual tracks are smaller, the track gets read completely in less time.

As you get towards the other side, it takes longer to get access to the data, the platters have to rotate until the data you want gets under the read/write heads. That's why the latency goes up.

The read/write heads don't move back to the side every time they're done reading something. The drive heads will go left and right as needed, it's not move all the way to the end, then start seeking again

 

You want to cache the executables and libraries (usually small, within 1-10 MB each) at the beginning of the drive, because it speeds up booting the operating system, when there's loads of files read, often in parallel. You want to seek as little as possible, and you want to have small tracks, to reduce latency.

The operating system will CACHE a lot of these executables in RAM one they're read and will only drop them if programs need free RAM memory. If you launch a game, play for a few minutes then quit and 30 minutes you launch the game again, the executable of that game is probably still in RAM and the mechanical hard drive won't have to move the heads to seek it on the partition and retrieve the data.

 

It's not about speed, it's about latency. The high throughput, those high speeds, are achieved by mechanical drives when they can read a big file, a long track... basically they put the read heads over a track and as the platter spins, they put that track in those 16/32/64/128 MB of cache on the drive and push it to you.

If you read loads of very small executables, the drive won't get the chance to speed up to its maximum speeds.

 

For video and audio files, it makes most sense to be put towards the end of a drive, because they're static, they're not changed unless you edit the ID3 tags or whatever. Therefore, they won't fragment in more pieces.

Also, the speed of modern mechanical drives is fast enough that it won't be a problem if the files are located at the end where the speeds are lower (ex 260 MB/s at the beginning for a 6 TB modern drive, but only 150 MB/s on the outer tracks - well, even a 4K video only needs 20-30 MB/s sustained to be watched)

You also don't care about latency, because if the files are not fragmented, the video player can simply read ahead a few hundred MB of the video file and keep it in RAM as a buffer. For example, Media Player Classic Home Cinema is configured by default to buffer up to 256 MB of data ... for a regular bluray encoded at 5-10 MB/s that's basically 3-5 minutes of video read in advance from a mechanical drive within 4-5 seconds (at 100+ MB/s sequential read)

 

As for video games, you're assuming that video games simply read sequentially from those big files. No, often they randomly search within those files and extract and load in RAM the parts they need. Just making a partition in the middle for games it's not enough.

I'll give you an example.. let's say you install 3 games on a 100 GB partition :  one is 35 GB and one is 35 GB and one is 20 GB.

When you install the first game, Windows will automatically pick random write locations for the files of that game :

So you install game one and the files are randomly placed in that small partition.

When you install the second game, Windows will search for the biggest empty areas and put the game files spread across the partition wherever there's big empty areas, so you could end up with something like this:

Now you're about to install the third game, but if the game has a 10 GB big file, there's no single empty space to put that single 10 GB in one big continuous area, so the file will become fragmented.

Maybe the 3rd game will be like this:

 

 

 

So even if you make your fancy pants partitions, from time to time, with a mechanical drive you have to defragment it, to make sure the files are continuous and that the hard drive gets a chance to read those files sequentially.

By making partitions, you're making it harder for the operating system to find big enough empty areas of drive space, to reduce fragmentation of big files.

In lots of games, those big files aren't really read sequential anyway... think of those big files like zip files and every time you load a game level, the game loads 10% of a big file, picking out hundreds of small files from random places out of those big files, or just a few files (depending if some files are already loaded in ram and can be reused). There's a big file for textures, a big file for music, a file for dialogues speech and sound effects, a few files for each game level information and so on...

 

You have some misconceptions about how things work so you THINK you're 100% correct, but in real world things don't happen like you think they happen.

 

 

 

 

 

[mariushm]

Also don't quote all my post. Please edit your posts and remove my quotes, it's just us two exchanging messages so it's obvious who replies to what person.

No point hurting your fingers scrolling.

 

Some other remarks...

 

You're drawing conclusions based on TWO SAMPLES, one 7200 rpm and one 5400 rpm.

You're not accounting for platter data density, you're not accounting for platter count and number of read/write heads, you're not accounting for cache size on the mechanical drives, you're not accounting for lots of things. 

A hard drive with 3 platters may have higher latency than a drive with just one platter.

A hard drive with higher platter data density (gigabits per square inch of platter surface) will behave differently than platters from an older generation.

 

Rethink everything accounting for caching of files and what I wrote above...

[wkdpaul]

* thread cleaned *

[mariushm]

30 minutes ago, Goildzy said:

*snip*

Explain where my own benchmark contradicts what I said.

I am almost 40 years old, I have worked in IT since I was maybe 15 years old and I have a lot of experience with hardware and I know how computer components work and even repaired a few of them. I repair electronics, I also program stuff, I know what I'm talking about.

In contrast, you're talking like a 10 year old who thinks he discovered something nobody else thought of, and draws conclusions using the numbers of a benchmark he doesn't understand.

For example, prove you're smart and know the minimum by explaining what that 1 GiB means in that combo box in CrystalMark and what Q32T1 means and maybe see if the benchmark results will continue to point to what you say if you change from 1 GiB to some other value, like 10 GiB

 

Also, you're a person that repairs "headers" ... what are you talking about, connectors, motherboard headers? You can't even write "heads"?  

Someone typically doesn't repair drive heads. Someone buys an identical drive, opens it up, moves the head assembly to the faulty drive and recovers data to a new hard drive. Problem solved. The service person's time often costs more than the price of a spare mechanical drive and no good repair person would waste time repairing individual heads.

 

You've also quoted my long post 3 times already even through I asked you not to.

 

 

Edited October 26, 2019 by wkdpaul

[Spotty]

- Thread Cleaned & Locked -

Quote

Encourage the freedom of expression and exchange of information in a mature and responsible manner.