• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Looking to run Linux

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Originally posted by: SleepWalkerX
Originally posted by: Nothinman
I found my answer when someone mentioned the fact that the raid that is installed is not technically a hardware raid and that performace on those drives are terrible. Which got me wondering why I installed Linux in the first place, because performance from the Mac machines to the server to the RAID was terrible. So I decided to test out what would happen if I transfered to a single disk. As it turns out, the raid performace is so bad that for my solution it warrants a hardware raid, and I will simply run windows off of it.

Software RAID shouldn't be any worse than a single drive, at least nothing noticable. If it is I would say something else is wrong in your setup too. Infact in lots of scenarios Linux software is faster than a hardware RAID controller.

That's what I thought.


Yep. That's right.

Some links:
http://linas.org/linux/raid.html

This one shows Linux software RAID vs Hardware RAID on the same controller. However I am told that it's a 3com controller known for lackluster performance.
http://spamaps.org/raidtests.php

When looking at the graphs entries that begin with HW are hardware and SW for software. SWEXT3 would be software raid on ext3.

Keep in mind that this benchmark is old. Peformance of software raid and file systems have improved.

Looking at the benchmarks with that software raid beat the hardware raid in every single instance. On especially favorable benchmarks the software raid outperformed the hardware raid by more then 200%.

There have been other benchmarks that I've seen and they've always shown software raid is faster.

The reason Linux software raid is faster is because of 2 things...
A: Linux MD (stuff in kernel responsable for software raid) is fantastic. Smart algorythms help. Laying out data by partition rather then abstract it with hardware has good performance advantages. (something about how the blocks are laid out. don't understand it myself).
B. The proccessors used to hardware raid are usually generic embedded items. They typically run around 200 or 400mhz. Unlike 3d cards they general purpose proccessors. Your 3000mhz Pentium 4 is much much more efficient and much much faster then those little things.

Nowadays it's much cheaper to spend the money on a computer with excessive amounts of proccessing power then it is to go buy a dedicated 'real' hardware raid. It used to be that servers would have their cpus heavily loaded so offloading I/O to seperate hardware made sense from a performance perspective. But nowadays cpus are very cheap and fast and getting a dual cpu box is cheaper then getting a single core + hardware raid.

The reason you still want hardware raid is features for other then speed. The hardware usually incorporates extra monitoring features and error correction/detection. It supports hotswapping. And things like that. It's simply more reliable and will help you protect your data and lower administration overhead. (generally speaking. Some stuff is just crap).

Also keep in mind that software raid has limited scalability. With the regular 32bit PCI bus it will run out of bandwidth after about 5 harddrives. (Of course with PCI express, however...)

Nowadays raid makers know this so they are starting to make fakeraid devices that incorporate some real hardware raid features. So it's become tricky to find 'real' hardware raid unless your going scsi.

This page is handy in my experiance..
http://linuxmafia.com/faq/Hardware/sata.html
 
Originally posted by: SleepWalkerX
Originally posted by: Nothinman
I found my answer when someone mentioned the fact that the raid that is installed is not technically a hardware raid and that performace on those drives are terrible. Which got me wondering why I installed Linux in the first place, because performance from the Mac machines to the server to the RAID was terrible. So I decided to test out what would happen if I transfered to a single disk. As it turns out, the raid performace is so bad that for my solution it warrants a hardware raid, and I will simply run windows off of it.

Software RAID shouldn't be any worse than a single drive, at least nothing noticable. If it is I would say something else is wrong in your setup too. Infact in lots of scenarios Linux software is faster than a hardware RAID controller.

That's what I thought.

If thats the truth, why can I transfer the same file from the Mac (3.21GB) to my system drive, and achive gigabit speeds easily, and transfer to the raid and the transfer takes 15 minutes? My boss and I decided to get a Hardware raid card the fact that we're doing video production makes speed very important. Software raid is extremely slow when it comes to write rates.

EDIT: We'll probably go with RAID 3
 
If thats the truth, why can I transfer the same file from the Mac (3.21GB) to my system drive, and achive gigabit speeds easily, and transfer to the raid and the transfer takes 15 minutes? My boss and I decided to get a Hardware raid card the fact that we're doing video production makes speed very important. Software raid is extremely slow when it comes to write rates.

Going from Mac using Samba onto a Linux server running Software raid?

There is a lot more going on in a network transfer then just RAID disk speed.

Although it could definately be software raid problem. With 6 disks + gigabit nic your going to run into PCI bus bandwidth limitations. A hardware raid card will help that out a lot.

Although there is a lot more going on in a network transfer then just drive speed. Also on different raid systems will take a long time to build up. For instance RAID 5 system will take several hours to build.. You'll be able to use it,but it will take a long time before it's fast.

If you still have your system up it will be worth doing benchmarks on teh file system to make sure you know exactly were the bottleneck is at. If it's the disks then the raid card will probably help, but if it it's a problem with the network or how SAMBA or whatever is setup then it would be a waste to spend a lot of money on something that won't solve your problems.

A simple benchmark for SAMBA performance would be a FTP transfer, for example. A properly setup SAMBA server should be within 60-70% of the same performance as a ftp transfer.


 
If thats the truth, why can I transfer the same file from the Mac (3.21GB) to my system drive, and achive gigabit speeds easily, and transfer to the raid and the transfer takes 15 minutes?

Without doing any real investigation I can't tell you, but just looking at it and saying "This RAID card must suck" is definitely not the right solution. There's a lot of variables involved in such large transfers, picking one at random isn't the best way to troubleshoot anything.

Software raid is extremely slow when it comes to write rates.

Not true at all. Software RAID 1 will be about the speed of the hardware, software RAID 5 will take extra time because it has to use the host CPU to do the xor calculations but it shouldn't be slow enough to justify a several hundred dollar RAID card unless you're doing tons of simultaneous rights. Just a guess off the top of my head, but I would say that you shouldn't need a hardware RAID solution unless you have at least 5 writers at once. And even then I would run some benchmarks to make sure it's really worth it.

EDIT: We'll probably go with RAID 3

Why? You essentially lose one drive worth of performance because you're putting all of your parity data there instead of striping it across all of the drives. Do you have any real reason to choosing RAID 3 over RAID 5?
 
Originally posted by: Nothinman

EDIT: We'll probably go with RAID 3

Why? You essentially lose one drive worth of performance because you're putting all of your parity data there instead of striping it across all of the drives. Do you have any real reason to choosing RAID 3 over RAID 5?

Raid 3 is alot better for Large files where write bandwidth is needed. Such as video streaming, which is exactly what we are doing.

I've done my research, I've tried plenty of senarios and tweaks. I've looked at numerous benchmarks, all telling me that RAID 5 write speed is slow, and that with a good hardware raid card can be very good. The card we are getting does raid 5 and raid 3, so once I get it I'll prove to you that it really is the bottle neck.

I know the bottleneck is at the PC, I've done tests where I'm teaming the gigabit ethernet controllers, and tried streaming from 2 computers at the same time, this causes the computer to drop both connections because it can't write to the hard drive fast enough.
 
Back
Top