Should You Enable Hibernation in Server 2003+

[LWComputingMVP]

(This is meant to be a DISCUSSION!)

I'm not having a problem with this per se, but I wanted to get a sampling of opinions. I suspect the consensus is NO!!!!, but I could be wrong. I do NOT typically do this, but when I find it enabled, I don't typically shut it off either. My line of thinking is this:

If you enable hibernation in a Windows Server 2003 or later system, then you can configure your UPS software to go into hibernation upon a power failure, in theory, resuming faster and POSSIBLY saving/preventing data from being lost.

One drawback is that hibernation CAN be flaky AND it CAN eat up disk space (the hiberfil file must be the size of RAM - on a 4 GB machine this could eat up 1/3 of the disk space - or more, depending on partition size).

Anyway, what do you all think?

 

[mrdenny]

That would be a no. We've got 64gigs of RAM in the servers here, and that would require like 90 Gig C Drives to do.

Denny
MCSA (2003) / MCDBA (SQL 2000)
MCTS (SQL 2005 / Microsoft Windows SharePoint Services 3.0: Configuration / Microsoft Office SharePoint Server 2007: Configuration)
MCITP Database Administrator (SQL 2005) / Database Developer (SQL 2005)

--Anything is possible. All it takes is a little research. (Me)

http://www.mrdenny.com

 

[LWComputingMVP]

Care to elaborate?

Fine - on servers with 64 GB of RAM, it may not make sense - the save time might be longer than the UPS battery... but if your DCs are running with 64 GB of RAM, then you almost certainly have wasted a lot of money. I find it doubtful (and at least wasteful) that all your servers have 64 GB.... so what about the others. What do YOU think the drawbacks are? Again, this is meant as a discussion - pros and cons - a little more than 2 sentences please...

 

[FloDiggs]

I understand your rationale, but I would still say no. The disk space is pretty valuable, and an environment where the extra few minutes of time is that valuable should probably invest in a backup generator of some kind.

 

[LWComputingMVP]

Fair point - and if you are NOT taking hibernation into consideration, your C: drive should not be very large... but if you do, then you might want to add 4-8 GB to the overall size at the beginning - still, Disk space is so cheap now... .25 per GB in a SATA system and even SCSI is around $1 per GB...

Obviously, this needs to be considered and not blindly done based on server config and what it's doing... but let's consider this: would you say it's UNWISE to do? Especially in small environments or on servers with relatively little RAM?

 

[itsp1965]

One thing you have to consider also is that with hibernation you are creating another large disk file that it will increase the amount of paging on your disk which will decrease performance (by how much -- that depends)

 

[LWComputingMVP]

Ok... maybe I don't understand this - but the hiberfil file is used only when the system goes into hibernation - so how exactly does merely having a large file increase paging?

 

[itsp1965]

True.. but isn't the data within the hibernation file always refreshed? A server is likely to have more data residing in memory at any time than a laptop, therefore once it goes into hibernation it has to update/overwrite the contents of the file with new data. Now depending on the usage of the server, this could happen a lot.
Also when the server is "woken up" from a hibernated state, doesn't the disk thrash during this process and it's also possible something is being written to the pagefile during this process.

 

[LWComputingMVP]

This is something that should be looked into. My understanding of hibernation was that upon being told to hibernate, the RAM is dumped to the hiberfil.sys file. Upon resume, RAM is read from that file. In THEORY (or at least as my logic works), the pagefile should be unchanged and not accessed from the point it is told to hibernate until the system has fully resumed.

 

[itsp1965]

On a similar note it seems that Windows 2003 does not support hibernation on servers with > 4GB of RAM

http://support.microsoft.com/kb/888575/en-us

Part of the reason as it indicates is that a larger hibernation file can lead to poorer performance

 

[LWComputingMVP]

Not surprised that 4 GB is the cut-off - though I wasn't sure if that cut-off would apply to 64bit versions as well.

The performance thing is definitely something that should be explored and better understood. Is it JUST because of technological limitations or is something that results in slower and slower performance the more RAM you have - in which case, though it's not directly part of this question, it makes one wonder if there is a similar performance consideration on XP/Vista systems that have hibernation enabled.

 

[itsp1965]

It seems that this applies to Windows XP also (makes sense for Vista also)
To me it makes sense, since as I have indicated imagine having a server with 16,32 and 64GB of RAM. If you go into hibernation mode.. it could take a long time to write to the file thus causing excessive disk writes (even for a well optimized disk subsystem).
When it comes out of hibernation you are reading from disk and transferring back to memory, although this is less of a burden on the system than a write, it does temporarily take a great deal of CPU and disk cycles.

 

[mrdenny]

As I'm a DBA, all the boxes in my realm have 64 Gigs of RAM, except for the old ones with 32 Gigs of RAM.

I'm not sure how AD would handle comming back from a hybernation. If you are dealing with a full data center power down, I'm not sure that having all your machines recovering that quickly is a good idea.

Last time I had to deal with a data center power down we did a very controlled startup since we needed things to come up in order. Network Switches, DCs, Exchange, SQL, App Servers, then backup servers, EDW etc. We were afraid that having all the machines power on at once might overdraw the amount of power that the UPS could give out since startup requires more power than running.

Based on all this I'd still have to go with not hybernating and doing a full shutdown. Mostly for the controller startup than anything else.

I can only imagine the havock of having app servers comming up before SQL, and before the DCs. And as the App servers have the least amount of RAM they would be the ones that recover the fastest.

Denny
MCSA (2003) / MCDBA (SQL 2000)
MCTS (SQL 2005 / Microsoft Windows SharePoint Services 3.0: Configuration / Microsoft Office SharePoint Server 2007: Configuration)
MCITP Database Administrator (SQL 2005) / Database Developer (SQL 2005)

--Anything is possible. All it takes is a little research. (Me)

http://www.mrdenny.com