• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

SSDs for NAS?

U

UnholyCarcass

Weaksauce
Joined
Aug 27, 2015
Messages
125
Is it viable to use SSDs in place of WD Reds or any other enterprise level drives in a NAS? Does the performance increase justify the cost or will the system's speed be bottlenecked by the onboard LAN or wireless connectivity?
 
You are going to hit network bottlenecks and increase your cost. I am not sure what you are trying to solve, but SSDs are not any more reliable than hard drives. Large Data center storage systems that use SSDs burn through them just like they do hard drives. Equipment fails.
 
Spindle disks like the WD Red offers a sequential performance of about 40-200 MB/s,
depends on file size or if data is on the inner or outer lane of the disk.
If you access such a disk over a 1 Gb network, the network performance is enough/similar for a single user
sequential transfer with a large file.

SSDs offer a constant sequential transfer rate of about 500 MB/s.
For a single user this is quite irrelevant compared to a disk.

The main difference is iops. While the best spindels offers about 100 iops, a good SSD can give you some 100000 iops even under heavy load. If you have more than one user or need random access to small data, the difference is huge even on a 1Gb network.
 
Last edited:
_Gea said:
Spindle disks like the WD Red offers a sequential performance of about 40-200 MB/s,
depends on file size or if data is on the inner or outer lane of the disk.
If you access such a disk over a 1 Gb network, the network performance is enough/similar for a single user
sequential transfer with a large file.

SSDs offer a constant sequential transfer rate of about 500 MB/s.
For a single user this is quite irrelevant compared to a disk.

The main difference is iops. While the best spindels offers about 100 iops, a good SSD can give you some 100000 iops even under heavy load. If you have more than one user or need random access to small data, the difference is huge even on a 1Gb network.
Click to expand...

The HDD IOPs are underestimated and the SSD IOPs are over estimated.

HDDs run from 200-500. 200-300 is most common range but some reach 500 (~2MBps)

SSDs 100K IOPs is blatant cherry picking.

10K is a more realistic number. Unless you are looking a a specific task. That 100K is pure dreaming.


SSDs make sense if data size is small to moderate and IOPs are crucial. If you need sub 10TB and you have lots of users SSDs are a possibility.
 
SomeGuy133 said:
HDDs run from 200-500. 200-300 is most common range but some reach 500 (~2MBps)
Click to expand...
15k rpm drives reach 250ops max, 7200rpm reach 120ops max, simply due to the rotational delay. Do the math.

SSDs 100K IOPs is blatant cherry picking.
Click to expand...
Random parallel 4k accesses is blatant cherry picking?
 
SomeGuy133 said:
10K is a more realistic number. Unless you are looking a a specific task. That 100K is pure dreaming.
Click to expand...

Per datasheet good SSDs are up to 100 000 iops
with a lower value of some 10 000 under load on writes. This is correct.

NVMe SSDs are up to several 100 000 iops per datasheet

Read iops are much higher than write values on any SSD.
 
Last edited:
_Gea said:
Per datasheet good SSDs are up to 100 000 iops
with a lower value of some 10 000 under load on writes. This is correct.

NVMe SSDs are up to several 100 000 iops per datasheet

Read iops are much higher than write values on any SSD.
Click to expand...
zrav said:
15k rpm drives reach 250ops max, 7200rpm reach 120ops max, simply due to the rotational delay. Do the math.


Random parallel 4k accesses is blatant cherry picking?
Click to expand...



my SSD also claims it'll do 550MBps or waheetver and that is ATTO and that is a useless measurement. -_-

https://www.dropbox.com/s/4ja1lsrcsqdf8h7/Samsung 840 500GB crystaldiskmark 1 to 4.png?dl=0

here is a real world example. You are dreaming to expent 100K IOPs unless you are Q32 and even then 100K maybe possible or maybe not.

first scan is it brand new with virtually no wear and the others are after it has seen some use and the speeds start to tank. It is even OP as well.

Do the math. I have owned many HDDs that do .5-2 MBps in 4K. Here since you want to talk like you know anything here is a fucking real world test so quite spreading bad info.

https://www.dropbox.com/s/v2ifsz68dae9hwk/crustal marc.png?dl=0
https://www.dropbox.com/s/c728qm73bwzxm9o/well this shows betteer wtf.png?dl=0

here read the damn real world results from a 1 TB 3.5 in spinner. Wow yes that's right 300 IOPS from a 1 TB that is from 6 years ago or something like that. There is also ~500 IOPs in writes 1.9MBps.
 
_Gea said:
Read iops are much higher than write values on any SSD.
Click to expand...
Low-queue random writes can be faster than reads, but I agree otherwise. Note that can apply only with a very small number of users.

SomeGuy133 said:
I have owned many HDDs that do .5-2 MBps in 4K.
Click to expand...
Note that Crystal creates a single test file, which is then accessed randomly. This is the ultimate example of short-stroking for performance gains. You'd see much lower results testing across the entire drive.
 
HammerSandwich said:
Low-queue random writes can be faster than reads, but I agree otherwise. Note that can apply only with a very small number of users.


Note that Crystal creates a single test file, which is then accessed randomly. This is the ultimate example of short-stroking for performance gains. You'd see much lower results testing across the entire drive.
Click to expand...

that doesn't really affect IOPs IIRC (could be mistaken). But either way his 120 IOPs MAX is total garbage. 500 is extreme end obviously and 200-300 is average. I have many different drives that show 200-300 is average. Not 120 being MAX that he is claiming. 120 is laughable and not even remotely close to being accurate to real world results and he is claiming that is the max wtf???
 
IOPS will be determined by the size and randomness of the input-output. Cant really compare across the board unless the parameters are the same.

On the subject of max IOPS "cherry picking"
Just like in a foot race, it doesn't matter what your technique is. just your overall time. Who cares if you swing your arms or shave your body?
 
This is a good read, and I am subscribed. I've always wondered if I could use an SSD inside of a nas. Problem is most bays are 3.5''. Well mine is.
 
ReCOde said:
This is a good read, and I am subscribed. I've always wondered if I could use an SSD inside of a nas. Problem is most bays are 3.5''. Well mine is.
Click to expand...

Well another issue which I do not think was pointed out in the thread yet is TRIM, I do not think most off the shelf NAS boxes support TRIM for SSDs which could cause issues, or maybe I am wrong here, Discuss!
 
zaniix said:
Well another issue which I do not think was pointed out in the thread yet is TRIM, I do not think most off the shelf NAS boxes support TRIM for SSDs which could cause issues, or maybe I am wrong here, Discuss!
Click to expand...

Nearly every OS supports trim for single SSDs but mostly this will not work with SSDs in a raid. In such a case you must rely to the SSD controller based built in garbage collection.

The quality of the internal garbage collection differ. Modern enterprise SSDs like Intel S3500 .. 3710 work quite well without OS trim support. The qualitly of the controller/ firmware is important as well as a large overprovisioning up to 40%. On consumer grade SSDs you can keep write performance higher if you add such an overprovisioning manually either by partitioning a new SSD or creating a hidden host protected area HPA with tools like hdparm or hdat2.
 
You must log in or register to reply here.
Back
Top