My set-up: any tips?

[Neglacio]

Hey guys!

First want to tell you that I love your software. I first learned from this kind of caching when my new SSD came with its own cache: Samsung RAPID, which does exactly the same, but doesn't allow for customization. It's just a 1GB cache for the SSD with a defer-write that can't be changed but may be optimized for this kind of SSD. Primocache performs a lot better in CDM but when the cache is almost full, Primocache gives up but Samsung RAPID stays stable. Some tests I made when I first started with Primocache, beware, it's messy: http://pastebin.com/iJTa9kA3
However, I'm just using Primocache ATM. I prefer to use more RAM for Primocache so it doesn't get full. Primocache offers more customizability than RAPID and a much better UI.

I'm currently using the following set-up, all read and write:
4GB RAM for 256GB Samsung SSD 840EVO (System)
2GB RAM for 4TB Seagate RAID0 HDD (Data)
256MB RAM and 256GB L2 cache for my external drives (2TB WD and 1TB Toshiba)

I have 16GB RAM in total, so I don't care about so much cache. I actually bought 16GB instead of 8GB to do some fun stuff

The 4GB for the SSD hits at around 20%. It's always around 60-70% filled. Is this a good ratio or am I doing something wrong?
The SSD is already fast enough on its own, but I prefer the caching for the defer-write. Writing larger blocks is better for the NAND chips, especially now the EVO has Triple Level Cells, which, in theory, should expire faster.
I don't see any direct benefits in-game or while browsing, except for the superfast file transfer and program installations.

The 2GB for the Data drive however hits at around 50% after a 5-6hour of full-bandwidth torrent transfer.
Is it useful to enable defer-write on a normal (fast) HDD? There would be no gain here I think.
Does using this cache (including differ-write) mean that there is less wear on the disk, so I could use my RAID0 for longer?

On my 2TB Western Digital there is no data ATM. I just bought this computer and moved all my data from the WD to my RAID0. Therefore I made it a shared cache with my 1TB Toshiba. It will be used as back-up drive.
My 1TB Toshiba external drive hosts my Truecrypt folder. In this is all my personal data (not getting into detail ), which I access frequently. The cache on this hits at 10-15% which I expected: too small of a cache and because of the encryption, Primocache can't figure out how to cache correctly. However, it does give me a speed boost much higher than the USB speeds.
The 256GB L2 cache is hosted by an old 2.5" Samsung HDD, which was an external drive itself back then. This too gives me a speed-benefit. I currently have no use for this small HDD, so I plugged it into my hotswapbays.
I'm not using it for the Data drive, because the Samsung HDD is slower (or am I wrong?!?) so it would give no benefit there.

Do you guys have any more tips on what I could do? I currently have 5GB of free RAM with my browser, emailclient, media player, ... opened up. When I do some gaming, I this drops to around 2GB of free RAM.
I would love to buy it when it's released, for a reasonable price ofc , but I'd like to have it at max settings before that

Many thanks!
-Neglacio

[InquiringMind]

Welcome to the forums Neglacio,

The reported cache ratio is the percentage of hits since system start, and will depend on your usage. 20% seems typical (running a benchmark like Crystal Diskmark will ratchet that up to 99% because of the large amounts of data CDM uses which can be cached and this will skew future readings until system restart).

Depending on your hard disk usage, write defer may be worth using (it risks data loss but you're already taking a risk with RAID-0 and presumably doing regular backups, right?). If the disk is used to hold applications and temp files then yes, if mainly downloads and media files (stuff that is written once, then read-only thereafter) then probably no.

For torrents, write-defer might offer a tiny performance benefit, but the main constraint is likely to be your Internet bandwidth - it's difficult to see write defer reducing hard disk wear much in that situation since torrent writes tend to be random.

Write-defer should be left off with TrueCrypt volumes (and encryption generally), since even a few bytes' data loss could result in larger sections of the volume becoming unreadable.

Two general suggestions with PrimoCache: it may be better to focus on the most used volumes (presumably the SSD and RAID in your cases) and allocate all the RAM you have spare to them rather than using it for drives that are rarely accessed; and experiment with blocksize - 0.9.8+ seems to offer a good setting by default but you may find larger (or maybe smaller) sizes work better.

[Neglacio]

All temp files are stored on the SSD. I used a ramdisk before, but I figured out a ramcache may be a better solution to reduce wear on the SSD.
If I understood this correctly, TRIM already happens in the ramcache, so it should be a huge benefit for the SSD.

I disabled write-defer on my data drive, since it was only used for torrenting. I notice no difference in performance, so I'll keep it this way. Thanks!
I used the default block size Primocache gave me, 64KB with 235MB overhead. Seems fine to me

All my external drives have a shared cache of 256MB L1 and 256GB L2. The L2 is a 256GB disk in my hotswapbay, which performs up to 2 times better than the USB disks. It also saves it content, so it doesn't have to learn again on a reboot.
While I was typing this, I noticed only 60GB of my L2 storage was used when I had much more reads and writes. So I disabled L1 cache all together, and now every read and write is cached on the L2 like it would have on L1, which could perform much better. I'm doing some tests ATM Too bad disabling L1 doesn't disable the memory overhead.
I disabled write-defer on my external drives before, since I didn't see any direct benefits here. Since they're less used, I can live with the extra wear, and I already read somewhere about the dangers with a Truecrypt container.

I regularly backup my complete C:-drive to a 2TB HDD. My (torrented) data isn't back-upped, but I might do so in the future. However, I already filled my 4TB data drive for 70%, so this might get expensive

EDIT: I just ran some tests with CrystalDiskMark, and I came up with the following numbers:


I use a larger size in CDM than the cachesize, but for me, this feels as close enough to a real-life situation. If the speed is worse, it means that the overhead is way too big. If the speed remains the same or better, I can only assume that any cache-hit will be at RAMspeed. Generally, for the Truecrypt drive, this is around 10%.
CDM is also one of the few benchmarks that can test a logical drive, like my Truecrypt one is.

As you can see, the 256GB HDD performs almost twice as fast as the 2TB Backup drive. I hoped Primocache would write first to the L2 cache, only to flush it afterwards, but it doesn't..
I'm currently sticking to just 256MB L1 RAM cache. Too bad, cause I can't find any other use for this drive

[InquiringMind]

All temp files are stored on the SSD. I used a ramdisk before, but I figured out a ramcache may be a better solution to reduce wear on the SSD.
If I understood this correctly, TRIM already happens in the ramcache, so it should be a huge benefit for the SSD.

If you have ramdisk software that works with Primo Cache, that would probably be a better choice for Temp files which (typically) need never be written to disk. It wouldn't make much difference to SSD lifespan though, most Temp files are small.

TRIM only comes into play with SSDs, not the cache (though it is technically possible for PrimoCache to align its writes to erase-write blocks, reducing the need for TRIM but I don't know if it does this). In practice, current SSDs seem to cope OK without TRIM (I've used WinXP without TRIM on SSDs for 3-4 years) thanks to background garbage disposal.

This Samsung SSD 840 Torture Test article should be of interest, since it suggests the 840s can last more than 3 times longer than Samsung claims.

I used the default block size Primocache gave me, 64KB with 235MB overhead. Seems fine to me

I've found the same too - PrimoCache does now offer a much better default size, but no harm in checking.

The L2 is a 256GB disk in my hotswapbay, which performs up to 2 times better than the USB disks...

Is that an SSD or mechanical hard disk?

...I noticed only 60GB of my L2 storage was used when I had much more reads and writes. So I disabled L1 cache all together, and now every read and write is cached on the L2 like it would have on L1, which could perform much better.

Given L1 is so much faster, disabling it would likely slow things down considerably.

I disabled write-defer on my external drives before...

A good idea since, unless you are careful to use "Safely Remove Hardware" every time before unplugging, it would be easy to lose data.

I use a larger size in CDM than the cachesize, but for me, this feels as close enough to a real-life situation. If the speed is worse, it means that the overhead is way too big. If the speed remains the same or better, I can only assume that any cache-hit will be at RAMspeed.

I hate to be the bearer of bad news but the speeds on your backup drive look rather low. A 2TB drive should give 120-150MB/s sequential read/write compared to the 35-40MB/s your screenshot reports for the backup drive. The cached screenshot showing lower figures seems surprising (if your 256GB drive was mechanical rather than an SSD that could explain it) - could you confirm that those screenshots are the right way round?

On the other hand, the Truecrypt drive seems surprisingly good for USB 2.0 (which should bottleneck at 40-50MB/s) - again have you got the cached/uncached screenshots the right way round?

The 256GB drive (which you say you're using for L2) is the key factor here - the CDM figures suggest it is mechanical rather than SSD, in which case you shouldn't use it for L2 caching at all (it will be slower than your 1GB/2GB drives). In that situation, just using L1 caching would be your best option.

[TELVM]

... This Samsung SSD 840 Torture Test article should be of interest, since it suggests the 840s can last more than 3 times longer than Samsung claims ...

Hmm, Hardware.info says that the first reallocated sector on a Samsung 840 TLC only appeared after 700 TiB.

However in a similar torture test The Tech Report experienced the first reallocated sector at 100TiB:



^ The MLC ones are still in good shape after 500TiB.

[Neglacio]

Well, let's hope Primocache will help increase its longevity

I recently did some big changes to my complete set-up.
The RAID0 lost its cache as it was ineffective and unnecessary. The 256GB HDD I would've liked to use as L2 cache has been promoted as backup-drive for my OS. The previous 2TB backup drive has been cracked out of its USB2 drive and connected with SATA, however, now I don't have any use for it I created a 2GB RAMDrive with software that came bundled with my motherboard, ASRock's XFast RAM, which holds my cache and temp files. I made it that big to hold temp archives and downloads until I do a complete reboot (which can be once a week, sleep ftw!).

Here's a list with my new setup, and some, as always, nice benchmarks:
256GB System Drive (C:) Samsung 840EVO SSD - 4GB cache (Read/Write, Write-Defer 60s)
256GB Backup Drive (U:) Samsung Spinpoint HDD - No cache
4TB Torrent Drive (D:) RAID0 2x2TB Seagate Barracuda HDD - No cache
2TB (Unused) Data Drive (H:) Western Digital Green - No cache
1TB Truecrypt Drive (P:, encrypted T:) - 1GB cache (Read/Write, Write-Defer 10s)
2GB RAMDrive (X:) ASRock XFast RAM - No cache


P.S.: I noticed my pc-store assembled the torrent drive with a software RAID0 whilst my motherboard is capable of RAID0. I guess that explains the low performance of random reads. I'm looking out for a used drive to backup my data, delete all partitions and try it again myself or disable the RAID altogether.

[TELVM]

Well, let's hope Primocache will help increase its longevity ...

Indeed. Anyway I wouldn't worry too much, for typical client use this SSD endurance stuff is technically interesting but largely academical. Even 100TiB tops would mean 11+ years of life writing 24GB every day, or 280 years writing 1GB every day.

[Neglacio]

Yeah, by the time that happens, there will be a new shining SSD with the speed of RAM available Even ARM joined the fight.
I'm already bumping against the small size of my 256GB SSD, at 10 games installed. I'm wondering: should I buy another 256GB 840EVO and RAID0 them or shall I wait another year and just buy a 500GB by then...
EDIT: And of course, Tom's Hardware made a review about this particular case http://www.tomshardware.com/reviews/ssd ... 85-13.html

[quigg]

Hi, I'm new to this forum and also a new user of primocache.

Like you, I've been playing for some days now with this incredible software and trying different settings. First, I tried separate caches for different drives/partitions, but I get by far the best results when I use one big cache for all the drives/partitions. At the moment, I'm using one 10Gb cache read/write (default delay of 10 secs) with block size 16, and the whole system feels very fast. To avoid problems with the psu, I also bought a cheap ups.

I appreciate any thoughts that anyone might have.

[InquiringMind]

I recently did some big changes to my complete set-up...

The new benchmark figures seem much closer to what I'd expect - hope the improvement shows in real-world usage too.

If you have a spare disk, then why not use it for more backups? (you can never have too many...). Assuming that you are making full image backups of your current disks (using the likes of Drive Snapshot, Acronis TrueImage, Macrium Reflect or something similar), the next step is having more frequent backups of your most important files (documents, savegames, etc).

File versioning software (which makes copies of files the instant they change) is a good complement to full backups - options include Aphar Backup (Dutch webpage but program is in English - my favourite since it is light and doesn't require .NET Framework), AutoVer, Yadis Backup (all free) and FileHamster (commercial).