Planned downtime: Harddrive replacement (again)

[BadMotel]

However, once the minimaps regenerate properly, I do have an experiment to run which might help with the lag.

Probably shouldn't have mentioned this. Now we're going to get 40 posts a day at least asking if you ran the experiment, when you're going to run the experiment, if you're going to again, what you're going to do, give stupid suggestions, and complain about lag the whole time throughout.

[GreenScape]

The minimaps not being there seemed to clear the lag up alot though, should just leave um off =s

I don't think so, though. I had considered the same thing, so I tried turning them off for a while a few days ago, but at that time it did not seem to clear up any lag.

Edit: If anything, I might actually imagine that the disk being down and therefore turning off RAID might have made the I/O situation better in itself. I have been suspecting the swraid to be partly responsible for some time.

I suggest using this shiny thingy. Software RAID controller. Very very good. On our servers we got 3x performance increase in compare to UFS and EXT*. Though you need FreeBSD. Just saying...

[loftar]

ZFS

Well, yes, I have actually considered switching to FreeBSD either way. From my previous usage of it, it is my impression that its IO-related subsystems are generally better than Linux'. However, it also means that there's quite a bit of code that needs to be ported and changed, and a server that needs reinstallation and reconfiguration, and time is not quite that plentiful. :)

I have also considered btrfs, but I'm not sure its stable enough for me to dare it yet. I've also let myself understand that it's having some trouble with BSDDB-like file usage. I don't know how ZFS is in that regard.

[tempwad]

seems like a plan. good luck.

[loftar]

It does? If so, I must have missed the actual plan. :)

[tempwad]

if you have no plan, take my, you know it: windows, c++, mt, tcp, zlib.

[GreenScape]

Well, yes, I have actually considered switching to FreeBSD either way. From my previous usage of it, it is my impression that its IO-related subsystems are generally better than Linux'. However, it also means that there's quite a bit of code that needs to be ported and changed, and a server that needs reinstallation and reconfiguration, and time is not quite that plentiful.

Yeah, i understand . But actually, if you were following strictly POSIX it's not that much of porting...

I have also considered btrfs, but I'm not sure its stable enough for me to dare it yet. I've also let myself understand that it's having some trouble with BSDDB-like file usage. I don't know how ZFS is in that regard.

I have no experience with BTRFS, so i can't tell something about it. Actually BDB is pain in the ass. It has a lot of cons. You should better consider Tokyo Cabinet and co. For example on FreeBSD there is a weird bug when your system runs out of memory and starts kiling processes while there is 90% of inactive RAM. We tracked this issue to our use of DBD. After we replaced it with sqlite - problem mysteriously gone. Not saying how slow BDB is, even after knowing how slow sqlite is .
About ZFS. It's really very good FS. Especially for mirrored data. Like you can take out HDD from raid in online and replace it with clean one. Very usefull for live servers when drives began to break. Great caching. Great usage of disk pool. When data is perfectly distributed over the pool when one file will be spread over several drives. etc.. etc...

[gt4329b]

things like "is the server up" and "when will the server be up"

If you use Chrome, try this: H&H Server Status Chrome extension - helps make the oh-so-long wait a little more bearable if you know a notification balloon will pop up when the server is online again

(H&H forum thread discussing this extension: viewtopic.php?f=7&t=16200)

[eXmart]

So.... loftar, what is your plan for the experiment? When are you going to do it?

[Zou]

he is going to take the toast out of the server