New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
It comes down to what setup you're using. RAID10 doesn't run bad at all, but RAID6 is pretty ugh. We got a 12 disk RAID6 in our storage array and it chugs. I think that's my fault for using the wrong chunk size when I made the box. Thankfully the newest versions of mdadm support changing the chunksize, it's just a very long process. I think it's going to take 1 1/2 weeks on our storage box <_<
Francisco
I personally prefer Raid5/6.
As @Francisco said, nothing's wrong with SW RAID-10.
RAID 1/0 can work fine on software, but it still puts unnecessary overhead on your load/io. On some systems that's not an issue, but on a vps node I don't see why you would ever choose software over hardware, since you don't want your resources doing things they don't need to be doing. Not to mention the fact that hardware RAID is almost always going to be faster (depends on the card).
My real problem with software RAID is when it fails, it fails hard. If a drive fails it can take forever to rebuild the array. During the rebuild, the node resources are going to be so overloaded, that you won't be running any vps containers until it's done. You might not lose any data, but the rebuild could take so long that you'll probably lose customers anyway.
The truth is, drives these days do not fail nearly as much as they used to. From personal experience, I can have a new drive installed and containers restored from a backup in a fraction of the time that it takes a failed software RAID to rebuild. It all just comes down to how you would answer this question: In the event of a hard drive failure, would you rather wait 24+ hours for a rebuild to finish and save all of your customer's data... or would you rather have the node back up within 1-2 hours, with restored backups? I've gone through both of those scenarios and I know which one I'd choose.
ps - It also really depends on who your datacenter is. My thought process is a lot different when I'm not using Wholesale Internet and Datashack.
Some DC's are just insane in how much they charge for a RAID card. RAID cards cost like $200 for a good adaptec 2405 retail, and 3wares they get for a dime a dozen.
So why does it cost $50/m for the card?
Francisco
Because addons and optionals let you regain what you had to cut on the server price to attract the client / fight the competition
People will pay that much for it, that's why.
I had various issues with software raid. Some were regular ones, like hdd marked as faulty when it wasnt, for example, others just mysterious fails when everything looked like ok, but performance degradation to sometimes KB/s and permanent rebuilds even on idle box, most of the time out of the blue. It was some time ago, tho, CentOS 3 times IIRC, it may be better today, but I keep hearing ppl having troubles with it. When I found out I am spending most of the time troubleshooting storage, I gave up software raid altogether and never touched it since.
Will put up a test box with 2 scenarios, raid 1 and 0, if that passes decently, including rebuilds under load (not on 0, of course), I may consider it again. I wont even consider raid 5+, no matter what, if I need raid 5 it is probable that the data/uptime/performance warrant HW adapter.
M
I will agree that most software raid is crap.
However, Linux mdraid is excellent, and performance is nearly identical to hardware raid. I would imagine the raid built into ZFS/storage pools is similarly excellent.
I would take Linux mdraid over any hardware based raid any day. It's one less piece of hardware to worry about. When raid cards fail it's a nightmare getting the arrays working again. With Linux mdraid I get a consistent interface, and I don't have to install proprietary crap into my core system to monitor it.
We have not seen this with Linux mdraid. We have ours setup to do an integrity check (basically a rebuild since it makes sure both drives are consistent) once a month, and it's fine since it runs at the lowest priority and slows itself down when any load presents itself.
This isn't 1999. CPU's are plenty powerful enough to handle the extra load caused from Linux mdraid.
Our rebuilds (with 1TB 7.2k rpm drives) generally take about 6-10 hours. Same as any hardware raid solution.
@ramnet which raid setups have you had issues with? I have seen more failed MDRaid's than HW Raids. I know Raid 5, 50, and 60 can be problematic, but I have not seen any issues with RAID-10 builds of any size.
Course I work right across the street from LSI so if I have issues I can always chat someone up
On the other hand, sshfs at gigabit-ethernet speeds to a software raid 1 with USB2.0 drives in unventilated enclosures will bring a 2x dualcore Xeon to it's knees.
There may have been some other factors involved though
Pretty much any fakeraid that comes with motherboards is crap.
The management/monitoring software for Adaptec raid cards is also a PITA to get working unless you use RHEL.
That and I have had drives die in hardware raid cards from various manufacturers that cause a kernel panic when a drive dies. I've had that happen enough times to not like hardware raid.
I've always stuck with 3Ware and LSI cards and keep up with firmware revisions to deal with issues that can crop up. They don't have to be powerful since there's no parity calculations involved like Raid 5 would present.
@damian yep, CPU having to process SSH packets at line rate
changing ciphers can help with the load, but can reduce security.