Server junkies -- little help with deciding on new BBS server?

My vote would be for the best, it just depends on revenues versus income...

Always go for the best, maybe you can even negotiate a better deal?

DD

 

There shouldn't be any performance degradation in RAID 1 especially for reads. You are only reading from one drive and the card writes to both drives simultaneously. You just lose the space on the second drive. I never experienced slow performance on software raid 1 but I haven't tested it extensively on linux. Maybe the linux software raid implementation just sucks. Historically, hardware RAID were recommended when processors were slow and utilization were high because it off loaded the process on another dedicated chip. However, now most servers are running at 15 percent utilization so this no longer an issue. Also, what you want running this process? A processor on a card that probably running at 400 mhz or a quad core processor running 2.8 ghz on your server. For this reason, EMC Clarion SAN arrays runs Windows XP as its embedded OS. It is essentially using the whole computer system as a big expensive raid card.

The reason I generally recommend software raid is because there is no proprietary meta being used. The big drawback with HW raid is that every RAID card manufacturer uses it's own method to store disk configuration so a RAID disk can only be used with that family of card. If that card fails, you have to generally find the exact same card or card family or you won't be able to read the disk. This is not a problem with software RAID.

Actually RAID 5 is slower with small number of drives on writes because the card or machine have to calculate to parity bit. The reason people use it is because you only lose one drive for redundancy and you start gaining performance as the number of drives increases. I would never use RAID 5 with less that 5 drives.

RAID 10 is the by far the fastest but you lose half your space. All applications that require huge amount of IOPS.

 

This is flat out wrong. Reads are exactly the same since you are only reading from one disk just like you were if you only had one disk.

 

I can't help you much on speed, but I bought a Dell (through Honeywells) set up almost identical to this, and it's incredibly reliable. (Mine had two differences: RAID 5 with 5 73GB drives and Windows 2003) It's in a kind of dirty environment, and I've been out there with it full of dust and it's still running great, and plenty cool.

 

Clutch:

To answer your question, no RAID 1 would not be inefficient on reads. However, depending on the ability (hardware, application, end-user usage) to push the data through, RAID 5 will be 'more' efficient for read intensive applications. I think you made the correct decision, as the bbs is write intensive.

wnes/rockbox:

Just to clarify my comments on RAID 1, yes, it's main purpose is for redundancy. Yes, I agree, the read performance isn't going to be any less than a single spindle, especially if the OS has the ability to split seeks. My main point was that for databases that are mainly read intensive, RAID 5 will offer you an advantage over RAID 1.

Also, with respects to hardware v software raid. For critical applications, you want to go with hardware raid. Why? Most of these cards come with battery backed cache. If you lose power, the cache will still have time to commit your transactions, ensuring a higher degree of data integrity. For small applications, sure software RAID is fine. For critical enterprise-class apps, no way.

 

Option 2

 

From what I know, a lot of this depends on the RAID 1 controller and the OS, but if it's reading each disk in the array in parallel and split seeking, it could be a bit faster at reading than having no RAID. RAID 1's main advantage is just mirroring more than performance. It provides a backup/mirror in-case of drive failure. In its simplest form, you're just slamming 2 disks in that end up with the same data on both (mirroring), but you have the space of 1 of the disks. There is a write performance hit with RAID 1 from what I recall, but I don't know if it's that bad.

I wouldn't bother with RAID 5. I know Microsoft recommends RAID 1 and RAID 1+0 for web server and database installations the last time I looked and many DBA's puke at RAID 5 due to db performance. I'm not sure how mySQL db's are affected, but I can't imagine it's much different. If one of the drives fails in a RAID5 array, you will definitely take a performance hit until it's replaced, and then you have to rebuld the failed drive as well.

 

Have you considered going with the non-RAID system, and having a hot spare running on a cheap second system?

You don't need RAID for performance. Those systems have enough RAM to where I'm betting the whole BBS can be loaded into memory. RAID is really not the disaster recovery dream people think, because lots of things can go wrong besides losing a disk.

 

This guy might know something...

 

Critical? I for one would not lose any sleep if the BBS power-cycled and DaDakota mysteriously lost two posts on his race for post-count supremacy.

Maybe it's gotten better recently, but most cheap hardware RAID implementations suck balls. You can't trust them with your data; they bitrot your file systems and bring down your machine over time. Usually software RAID is faster and more reliable on the low-end.

Personally, the only good experience I've had with RAID is on SAN/NAS products. Everything else, I'd rather mirror the whole system than just a drive.

 

Like several others I would choose Option 2. Before you get into the best RAID level for the job it's important that you have at least some RAID level (above 0) to help protect against a disk crash. Still need to do backups because there are a ton of other things that could go wrong besides a drive crapping out.

RAID 1 vs. 5, 10K vs. 15K is a hard call without good baseline data on what kind of IO this BBS has. Someone earlier mentioned that this board was probably write intensive. I would think that it's more read intensive because my guess is that many, many more people read the posts than actually post and even the people who post read way more than they write. But that's just my guess.

So, Option 2 sounds like the most reasonable choice price vs. performance wise and, if you have room in the case and sufficient cooling, you can always add a raid card and three additional drives later to get a RAID 5 data partition. This would probably be the best scenario where you have the OS on a pair of mirrored drives and your data on a set of RAID 5 drives.

 

1

 

IMO the BBS is read-intensive on uneventful days, but can be very write-intensive during game time. This brings into question whether RAID 5 is a suitable implementation for the BBS. As has been said before, RAID 5 is almost as good as RAID 0 on read, but its write performance is poor. The other big drawback with RAID 5 is the significant performance degradation in the event of disk failure(s), to the point that the fault tolerance claimed by RAID 5 can be a joke.

Back in the days hard drive cost used to be a major hindrance to system upgrade, so RAID 5 was a cheap, reasonable alternative to RAID 1+0 in not-so-write-intensive applications. But if you are offered a RAID nowadays, you probably want to go straight to RAID 1+0 because hard drives are supposed to be a lot less expensive.

Really, it all depends on the kind of philosophy of operating this BBS. Life still goes on when DaDakotas of the world jump up and down in agony while the BBS is taken offline for maintenance.

 

I know I am late to chime in but would like to share some of my experience and see if you found relevant.

I used to mirror one of our servers (Raid 1) and one day we found the o/s was corrupted, unluckily same corruption was "mirrored" to the other hard disk hence it was a total failure. So if "mirroring" was the option, you may not want to mirror your o/s and applications. Those are relatively "static" data and recoverable via secondary media.

I encountered more than once hard disk crash and it is not even funny. If I have the option, I will go for RAID 5.

wnes, I believe your justification regarding the hard drive cost valids in a multiple disks environment. I think Clutch's case here is very likely a single disk environment, hence 2 hard drives for RAID 1 while at least 3 hard drives for RAID 5. So in his case, RAID 1 will be cheaper but RAID 5 will be more reliable.