Speed Questions with Linux Raid-1

[Netopia]

Greetings All!

Years ago, some hardware company (I don't remember which) came out with an on board controller that did what they named "Raid 1.5". The idea was that it created a Raid-1 mirror, but for reads it would read from both disks, getting chunks of data from each, giving read performance closer to Raid-0. GREAT IDEA.... but it never caught on in Mobo chipsets.

Today I was looking to find out whatever happened and stumbled across several posts that said that in current Linux Raid-1, reads are done like on Raid-0 and writes are done to only one disk until the system has time to sync the disks, thereby increasing writes speed to Raid-1 arrays.

But I found other posts that said this wasn't true about Linux software Raid-1.

So does anyone know the real deal? I'm going to be putting a box together for storage. I could get ~4TB with 3*2TB in Raid-5, but if Raid-1 in Linux actually has the performance advantages that some say, I might just get my ~4TB by using 2*2TB.

SO.... Anandtech Linux GURUs.... what say ye?

Joe

 

[Nothinman]

Yea, the Linux RAID1 driver does read balancing.

 

[Netopia]

Does that mean that as long as I'm using mdmda I'm ok? Assuming a single drive as the OS drive, what would be the best way to build the array?

Also, opinions about Raid-1 vs LVM?

Joe

 

[Nothinman]

Yes, mdadm is used to manage Linux software RAID.

Software RAID and LVM are separate layers, you can create a software RAID array and then use LVM on top of that. There is some overlap in the functionality, but I don't think many people use those parts of LVM. They just add a RAID array as a physical volume, add it to a volume group and then create logical volumes from it.

 

[PreferLinux]

I couldn't be bothered with LVM. I initially went with that because I only had Linux, but was wanting to be able to change partitions easily (which it claimed was possible). When I got Windows, I somehow had to resize the PV. That was a nightmare. Boot to the rescue CD, resize the file system, resize the LV, delete the PV, re-create the PV with the same start location but an end location significantly after the end of the LV. D:

 

[Nothinman]

I couldn't be bothered with LVM. I initially went with that because I only had Linux, but was wanting to be able to change partitions easily (which it claimed was possible). When I got Windows, I somehow had to resize the PV. That was a nightmare. Boot to the rescue CD, resize the file system, resize the LV, delete the PV, re-create the PV with the same start location but an end location significantly after the end of the LV. D:

That's because you were creating a new partition from that LV, if you didn't have to mess with the PV it would've been simpler. Although it's still not flawless, you still have to deal with the filesystem on the LV and stuff. It does make some things simpler but it also makes others more complicated since it requires a little more planning up front. It's a tradeoff.

 

[joetekubi]

Does that mean that as long as I'm using mdmda I'm ok? Assuming a single drive as the OS drive, what would be the best way to build the array?

Joe

If you've just got one drive at present, you will need to boot a live distro, backup all your partitions and data, change one or more partitions to Linux-raid, add a disk, create your raid set, and restore your data. Sounds onerous, but it isn't really. I've done it several times as I've moved from Raid-1 sets with 50 GB partitions to my current set with up to 500 GB partitions.

 

[PreferLinux]

That's because you were creating a new partition from that LV, if you didn't have to mess with the PV it would've been simpler. Although it's still not flawless, you still have to deal with the filesystem on the LV and stuff. It does make some things simpler but it also makes others more complicated since it requires a little more planning up front. It's a tradeoff.

Yes, I understand that.