• 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.

RAID general benchmarks

E

enderzero

n00b
Joined
Jun 28, 2005
Messages
4
Hi all -

Can anyone point me towards some general RAID benchmarks.

I am trying to see what kind of R/W performance numbers I can expect from the following setups.

Can anyone just tell me what order the following would go in for transfer averages.

1. Single SATA Hard Drive
2. 2 SATA HDDs Striped RAID 0
3. 2 SATA HDDs Mirrored RAID 1
4. 3 SATA HDDs RAID 5
5. 4 SATA HDDs RAID 5

I would assume the transfer averages would rank fastest to slowest 2, 5, 4, 1, 3. Does anyone disagree?
 
It depends on your drives, what type of RAID you use (Fake, soft, hard, etc), what level of RAID (0, 1, 5, etc), and how many disks you use.

RAID 0 is bar none the fastest of the RAID levels, but it is also the most unstable.

Mainly, it depends on the drives more than anything though.

Do a google search for some benches.

Do you want to do RAID yourself? What's the purpose of this?
 
The controller you are using is also a very big consideration, some controllers are much better than others.

Your analysis seems about right, although there is one thing that comes to mind that may affect RAID 1 performance. With RAID 1 you have the same data on 2 drives (or 2 stripes with RAID 10), some controllers are more intelligent than others and can read the data from both drives simultaneously, giving you read performance closer to that of RAID 0.
 
OldSchool said:
The controller you are using is also a very big consideration, some controllers are much better than others.

Your analysis seems about right, although there is one thing that comes to mind that may affect RAID 1 performance. With RAID 1 you have the same data on 2 drives (or 2 stripes with RAID 10), some controllers are more intelligent than others and can read the data from both drives simultaneously, giving you read performance closer to that of RAID 0.
Click to expand...

RAID 1 with FakeRAID (mobo) and hardware RAID will have increased read speeds by the number of disks x the original read speed.

Software RAID however may or may not have the increased read speed in RAID 1. It really depends on the OS/software and what algorithm is used. As far as I know, there is no way to increase the read speed in most Linux distros with MDADM, but there is a way in Open/FreeBSD using GMirror if you change the algorithm.

There may be a way to increase it with MDADM, but I haven't been able to find out how.
John4200 may know of a way how to do so though, you might PM him.
 
danswartz said:
ZFS does it right by default.
Click to expand...

It takes a lot of high-end hardware and memory to operate and can be very expensive however. It really can't be compared to RAID levels since it is not a true RAID system.
 
Red Falcon said:
It takes a lot of high-end hardware and memory to operate and can be very expensive however. It really can't be compared to RAID levels since it is not a true RAID system.
Click to expand...

What do you mean by high end with lots of memory? I run it on dual-core Optys with 2GB of RAM, and the data set is highly redundant.
 
LightningCrash said:
What do you mean by high end with lots of memory? I run it on dual-core Optys with 2GB of RAM, and the data set is highly redundant.
Click to expand...

What environment are you running that system in? How many people are accessing the data?

For a simple VM or data backup system, that would work fine, but it might struggle under heavier loads.

ZFS is very CPU and RAM heavy, depending on the environment it's running in.

This is taken from the ZFS thread:
Cheap and low power Intel Atom or the new HP Proliant Micro Server N36 (4 HD bays, max 8 GB ECC Ram, pcie 16x und 1x slots) with 1 GB+ RAM are ok but are running with reduced performance. But you should not use some goodies like deduplication, ZFS Z1-3 or encryption or it will become really slow. But it is suggested to add as much RAM as possible.

If possible, I would always prefer server-class hardware best are the X8 series from SuperMicro with 3420 or 5520 chipset like a micro-ATX Supermicro X8-SIL-F. It is relatively cheap with i3 CPU, fast but low power with L-Class Xeons (40W Quads), can hold up to 32 GB ECC Ram with fast Intel Nics and 3 x pci-e 8x slots and ipmi (browser based remote keyboard and console access with remote on/off/reset - even when ih power-off state). It also supports vti-d, so you may use it as a ESXi virtual server with pci-passthrough to a embedded, virtualized Nexenta or Solaris-based zfs storage server.
read about that mainboard http://www.servethehome.com/supermic...v2-diy-server/

minimum is 4 GB Bootdrive + 1 GB RAM + 32 Bit CPU (Nexenta)
best is 64 Bit Dual/ QuadCore, 8 GB ECC Ram+ (all used for caching)
If you want to use deduplication, you need 2-3 GB Ram per TB Data
Click to expand...

This is why I don't run it on my server and instead use software RAID. Until I can afford higher-end hardware with ECC memory, I'll stick to Linux distros and VMs.
 
Red Falcon said:
What environment are you running that system in? How many people are accessing the data?

For a simple VM or data backup system, that would work fine, but it might struggle under heavier loads.

ZFS is very CPU and RAM heavy, depending on the environment it's running in.

This is taken from the ZFS thread:


This is why I don't run it on my server and instead use software RAID. Until I can afford higher-end hardware with ECC memory, I'll stick to Linux distros and VMs.
Click to expand...

SMTP, Clamav, and POP3, 500 users, about 400-500 merged ops/s.

dedup obviously leans on RAM very heavily but other than that ZFS is great all around. Performance is only moderate compared to UFS or UFSIO but I've been very happy with it.
 
LightningCrash said:
SMTP, Clamav, and POP3, 500 users, about 400-500 merged ops/s.

dedup obviously leans on RAM very heavily but other than that ZFS is great all around. Performance is only moderate compared to UFS or UFSIO but I've been very happy with it.
Click to expand...

Sweet, that's good to know that it will work with that hardware in such an environment of heavy usage. Maybe I won't have to spend 2 grand on a ZFS system. :p

Thanks for sharing.
 
Red Falcon said:
Sweet, that's good to know that it will work with that hardware in such an environment of heavy usage. Maybe I won't have to spend 2 grand on a ZFS system. :p

Thanks for sharing.
Click to expand...

400-500 ops isn't really heavy, IMO.
You just have to get the right number of spindles for your workload.
I put 8 10Ks over from the JBOD, seems to be doing ok.
 
You must log in or register to reply here.
Back
Top