[NTLUG:Discuss] Software RAID NAS Box

[Daniel Hauck]

Interesting.  So swap space should be RAID too.

Richard Geoffrion wrote:
> Daniel Hauck wrote:
>> <snip the background to reduce the message burden on the list>
>>
>> Initially, I'm just thinking of mimicking the previous scheme when it was running under Win2K.  In this case, each drive will have the same partition scheme:
>>
>> 100MB /boot,
>> 2GB {swap},
>> 4000MB /
>> *everything else* /data
>>
>> The non-swap partitions should be RAID
>>
>> <snip some more>
>>
>> Okay, so I'm soliciting comments and better ideas..?
>>   
> Um, Daniel, 
> 
> TERSE VERSION:  non-raided swap space puts the server at risk of a crash 
> should a drive containing swap space fail.
> 
> VERBOSE VERSION: (in story form--should you choose to read the 
> "fictional" {ahem} account that I spent entirely TOO much time writing.)
> 
>  From the pages of that horror story "My server crashed when a disk 
> failed" let me recount the climax where the network administrator 
> realizes too late the perils of non-RAIDed swap space.
> 
> ---------------
> "I though you said the system would survive a crash if a hard drive 
> failed?  Isn't that why I bought multiple drives for my server?", 
> inquired the customer.
> 
> "It should have.", mumbled the network administrator, "It should have 
> survived the hard drive going down.  As a matter of fact, evidence in 
> the logs point to a memory failure. Only, I ran a RAM test and the RAM 
> tests clean."
> 
> Sudden realization filled the network administrator with dread - the 
> mistake of good intentions realized. The mistake wasn't just on this 
> failed server but on many of the servers.  Time bombs were ticking away 
> in the server room. The network administrator's mind raced to put words 
> to the issue.  Reluctantly the network administrator informed the 
> customer as to the cause of the crash.
> 
> "In a sense, RAM is hard drive space because hard drive space is USED 
> like RAM via the swap files.   When the hard drive went out, it took the 
> swap file with it and thus the memory pointers of the processes that had 
> been swapped out at the time.  It was in effect a lobotomy for the 
> server, hence the crash?"
> 
> "Well, what needs to be done then?  Is using multiple drives useless?"
> 
> "Using multiple drives still gives you protection from drive failure, 
> but we will want to RAID our swap partitions to protect the data they 
> hold from a future drive failure."
> 
> "But won't that reduce the amount of swap space we have allocated and 
> put a burden on our servers by reducing the memory available?" the 
> customer asked in concern.
> 
> "Technically, yes.", mused the network administrator, "But the need for 
> all that swap space isn't as critical in these days of increased 
> physical RAM.  Except for tempfs.  We can no longer use tempfs for our 
> /tmp directory.  Tempfs uses RAM -and thus swap space- to create a 
> temporary filesystem like a RAM disk.  We'll need to be sure to deal 
> with mounting /tmp in an appropriate place.   It looks like I'll be 
> spending some time in the server room correcting these issues."
> 
> "Better you than me." said the relieved customer.  "I'll be thinking 
> about you this weekend while you're working."
> ---------------
>