TLER + software RAID?

I

iroc409

[H]ard|Gawd
Joined
Jun 17, 2006
Messages
1,387
i am looking at maybe getting a bunch of WD2500JS drives (cheap!), and am looking to set up a gigantic RAID 5(ish) array. does the TLER error affect software RAID in Linux or FreeBSD, or is this merely a hardware controller error?

also... has anyone ran into any limitations with the software raid; i.e. is using 10 drives a problem?
 
I haven't seen any reports one way or another. For this reason, I tend to shy away from either family of WD drives with software raid.

My brother is running an 8-disk raid 5 in Linux. It works fine.
 
OK, thanks for the info. i haven't seen anything about it either, but i thought i'd ask.
 
just a bump... still looking for info. i will probably end up testing this myself, but would like info if anyone has it.

thanks.
 
From my understanding of TLER, the problem that WD was having is that controllers declare a drive as failed if it does not respond for x seconds. The TLER feature limits the time that a drive performs error correction to less than x (IIRC x ~= 5).

For all I know, the Linux's RAID implementation is an open source project. You could search the source code to see if there is a cutoff time during which drives need to respond in order to be considered `alive'.

If you do go through that trouble, it would be beneficial to the community here, if you posted your results.
 
i'm not much of a programmer, other than a little dabbling years ago. i'm not sure i'd be much good with that.

however, i was thinking i could set up a test to see if it would work. what would be a good way to do this; is there any specifics i should go over, or just set up a raid and let it run and see if it complains?
 
I run a mix of WD3200JD and RE drives in R5 on my Highpoint 2320 with no issues(1 year and counting). Just a FYI.
 
I know some fans of software RAID are going to hate the next thing I am about to say but here it is anyway: Non-hardware RAID sucks. Especially software RAID 5 where the CPU has to make all the parity calculations with the use of software. I've seen software RAID in action plenty of times and it's worse than the difference between hardware and onboard audio in terms of CPU usage. With a real controller with dedicated I/O processor you get FAR better performance. Plus I can take the controller and pop it into another machine much more easily than I can import software RAID disks into another machine.

That's my opinion.

LetFlameBegin.jpg


Like anything else in this business you get what you pay for and cheap RAID is RAID done wrong in my opinion.
 
Sorry, I can't let that lie :p
Dan_D said:
I know some fans of software RAID are going to hate the next thing I am about to say but here it is anyway: Non-hardware RAID sucks. Especially software RAID 5 where the CPU has to make all the parity calculations with the use of software. I've seen software RAID in action plenty of times and it's worse than the difference between hardware and onboard audio in terms of CPU usage.
Click to expand...
That's implementation-dependent. Actually doing the raid 5 calculations is ludicrously cheap - a pentium 3 can do around 2 gigabytes per second of raid 5 or 850 MB/s of raid 6 calculations - but if the implementation spends a lot of time polling for I/O or whatever it is they're doing, it can certainly suck up the cycles.
Dan_D said:
With a real controller with dedicated I/O processor you get FAR better performance.
Click to expand...
That depends. If you bought a raid controller based on the i960 chip from Intel (remember those?) you'd probably get owned by software raid, simply because it's so slow. So if you only have a hundred bucks to spend on a controller for 8 disks, you're better off not buying hardware raid because you'll end up with old, crappy, outdated hardware that software raid could outperform.
Dan_D said:
Plus I can take the controller and pop it into another machine much more easily than I can import software RAID disks into another machine.
Click to expand...
That depends entirely on the implementation. With Linux Software Raid, you can move it from machine to machine, put it on a different controller, and so forth, without breaking the array. I haven't tried moving it to a different architecture... but ZFS supports that. If you're using an onboard controller, it may be quite difficult to move it to a different machine or one with a different controller.

In short, what your impression of software raid is very dependent on what hardware raid and what software raid you've played with. If you get a chance and have supported hardware, check out Solaris 10's ZFS capability. It's pretty fast even on the awful, crappy hardware I'm using it on right now (can you say 33 mHz IDE bus?).
 
Dan_D said:
I... With a real controller with dedicated I/O processor you get FAR better performance.

Like anything else in this business you get what you pay for and cheap RAID is RAID done wrong in my opinion.
Click to expand...
If you are limited by the network (Fast Ethernet) anyway, who care about how fast your cheap-ass (TM) file server can read/ write files. We all know that 12MiB/s is pretty simple to achieve even for poor software R-5 implementations.


My Linux software RAID based server can send around 90MiB/s to my client via FTP over a GbE network. Upload around 40MiB/s.
 
drizzt81 said:
If you are limited by the network (Fast Ethernet) anyway, who care about how fast your cheap-ass (TM) file server can read/ write files. We all know that 12MiB/s is pretty simple to achieve even for poor software R-5 implementations.


My Linux software RAID based server can send around 90MiB/s to my client via FTP over a GbE network. <testing upload now>.
Click to expand...

True.

unhappy_mage said:
Sorry, I can't let that lie :p

That's implementation-dependent. Actually doing the raid 5 calculations is ludicrously cheap - a pentium 3 can do around 2 gigabytes per second of raid 5 or 850 MB/s of raid 6 calculations - but if the implementation spends a lot of time polling for I/O or whatever it is they're doing, it can certainly suck up the cycles.

That depends. If you bought a raid controller based on the i960 chip from Intel (remember those?) you'd probably get owned by software raid, simply because it's so slow. So if you only have a hundred bucks to spend on a controller for 8 disks, you're better off not buying hardware raid because you'll end up with old, crappy, outdated hardware that software raid could outperform.

That depends entirely on the implementation. With Linux Software Raid, you can move it from machine to machine, put it on a different controller, and so forth, without breaking the array. I haven't tried moving it to a different architecture... but ZFS supports that. If you're using an onboard controller, it may be quite difficult to move it to a different machine or one with a different controller.

In short, what your impression of software raid is very dependent on what hardware raid and what software raid you've played with. If you get a chance and have supported hardware, check out Solaris 10's ZFS capability. It's pretty fast even on the awful, crappy hardware I'm using it on right now (can you say 33 mHz IDE bus?).
Click to expand...

Again I agree in part. A cheap sub $100 RAID controller will deliver fairly poor performance compared to a higher end controller. Older controllers or those based on older processing engines are not going to be better. On those points I agree completely. As for software RAID, I never have messed with Solaris, but I haven't liked it in Linux. I also realize that the actual CPU utilization will be quite low just as is the case with software audio solutions.

I've always preferred having a BIOS and hardware to handle RAID levels of any type. It's always been a personal preference of mine and I don't see that changing any time soon.
 
it's good to know some people are having good luck with the software RAIDs and the WD drives.

dan, i understand your view on controllers; i think in an enterprise environment it's absolutely necessary. in my case however, i have tons of horsepower with nothing to torque on. the server i am building is for file sharing, and will run a handful of vmware systems (a couple xp boxen, a win server and a freebsd box, most likely). i will probably be running a 955 presler CPU with 4GB RAM... i'm not particularly concerned about running out of juice. we're also only talking about a max of a handful of users at once - and generally speaking only one.

i am actually seriously considering just running JBOD instead. i have not decided yet, but a RAID system may ultimately not really even serve me in any particular extra capacity. i think the redundancy is nice, but it all comes down to what my backup strategy ultimately becomes.
 
You must log in or register to reply here.
Back
Top