[Asterisk-Users] Hard drive write cache

Tue Jun 13 11:04:11 MST 2006

If all you are worried about is the write cache on the disks, why not 
just put the system on a UPS set to shutdown the system in the event of 
power failure, then place both the UPS and asterisk servers in a locked 
rack.  In the event of a power failure (or someone knocking the plug 
loose, which you can use locking plugs to further mitigate), the system 
will stay up on battery power then shut itself down to prevent data 
corruption.  I doubt you will get that level of uptime, but there are 
other options to help achieve higher reliability.  You can run the OS 
and asterisk on a solid state disk, and have voicemail and whatever else 
you want to go to rotating disks.  That will also help with power usage 
on the server when using the UPS.  Industrial flash disks are said to 
have (but they really can't promise this) a 3 million hour MTBF.

Thanks,
Nick

shadowym wrote:
> The cold hard truth is that if Asterisk cannot achieve 99.999% uptime
> without becoming much more expensive that a traditional PBX then it is not a
> viable alternative.  Even elcheapo Key systems are rated for five nines.
> That is what the telco world requires unless your just using Asterisk in
> your basement as a hobby or as a one man company.
>
> Redundant Servers is moving into the realm of non-competitive with
> Traditional PBX IMHO.
>
> I don't care about corruption of the CDR or any of the logging/database
> information.  All I care about is the ability make phone calls after power
> failure.  That IS the MAIN function of a PBX.  Not call centers, databases,
> CDR, click 2 call, and all the other bells and whistles.
>
>  
>
>   
>> -----Original Message-----
>> From: Boris Bakchiev [mailto:boris at jildent.com.au] 
>> Sent: Tuesday, June 13, 2006 2:13 AM
>> To: Asterisk Users Mailing List - Non-Commercial Discussion
>> Subject: RE: [Asterisk-Users] Hard drive write cache
>>
>> These days you don't have to worry much about your write 
>> cache unless you're running application where once single 
>> byte changed will affect whole file.
>>
>> Look at it this way, the only corruption will occur is 
>> whatever the files were open by asterisk at the time of the 
>> crash. And only up to the point where the file was last open. 
>> As far as I know asterisk does not keep cdr or log files open 
>> so you would loose only the data that was written at the time 
>> of the power failure.
>>
>> Any journaling file system (ext3, resierfs, xfs, etc) will 
>> easily handle any power failure event. Your files will not be 
>> corrupt but could miss some of the data.
>>
>> At the most you will loose 10-50 cdr entries written to you log files.
>>
>> If you post CDR to a remote SQL database then you asterisk 
>> install and linux is more or less static and will not be 
>> affected by the power failure.
>>
>> What you need to do is minimise the writes to hard disk's:
>>
>> 1 - Send syslog to remote server and do not do ANY syslogs
>>     Or keep the circular buffer in memory if you have plenty of it. 
>> 2 - Send CDR's to SQL server (or log to ramdisk and send to 
>> remote server every few minutes via SSH)
>> 3 - Do not record any calls (or do that somewhere else)
>> 4 - Stop any services that write/read data on regular intervals.
>>
>> If you have no writes you have nothing to worry about during 
>> power failure and journaling file system will take care of the rest.
>>
>> Keep your partition size really small so that fsck will not 
>> take much time.
>>
>> You have to be realistic, you cannot achieve 99.999% uptime. 
>> That's 5 minutes per year downtime.
>> You will have more or less 100% until your first hardware failure.
>>
>> Even if you have all the hardware components pre-purchased it 
>> will still take you 2-12 hours to detect, diagnose and fix 
>> the fault if you lucky.
>> So your 5 minuets 
>>
>> If the business is demanding 99.999% then it should be 
>> prepared to invest into the hardware.
>> I would recommend a cluster or even better a fault tolerant server.
>> Those are expensive but you can pretty much rule out the 
>> hardware failure and swap all of the failed components while 
>> the system is running (cpu, memory, hdd, etc).
>>
>> Look at Stratus or NEC FT servers if you need hardware redundancy.
>> They're expensive but will give you the hardware reliability you need.
>>
>> Or get a traditional PABX :)
>>
>>
>>
>>     
>>> -----Original Message-----
>>> From: asterisk-users-bounces at lists.digium.com 
>>>       
>> [mailto:asterisk-users- 
>>     
>>> bounces at lists.digium.com] On Behalf Of shadowym
>>> Sent: Tuesday, 13 June 2006 10:34
>>> To: asterisk-users at lists.digium.com
>>> Subject: [Asterisk-Users] Hard drive write cache
>>>
>>>
>>> I am looking at ways to harden my asterisk install to 
>>>       
>> prevent computer 
>>     
>>> related issues from happening.  I am concerned about about 
>>>       
>> disk write 
>>     
>>> cache.
>>> That seems to be a major source of hard drive corruption on power
>>>       
>> failure.
>>     
>>> Hard Drive corruption is simply unacceptable for the 99.999% uptime 
>>> requirements of my Asterisk install that needs to be as 
>>>       
>> reliable as a 
>>     
>>> proprietary PBX.
>>>
>>> Of course I will be using redundant power supplies, raid 1 and use a
>>>       
>> UPS.
>>     
>>> None of those things mean much if the power cords accidentally get
>>>       
>> pulled
>>     
>>> from the back of the server.  Unlikely as it may be I have 
>>>       
>> to consider
>> ALL
>>     
>>> possibilities.
>>>
>>> So is disabling the write cache a good way to reduce the 
>>>       
>> risk of hard 
>>     
>>> drive corruption for an Asterisk server?  I am not too 
>>>       
>> concerned about 
>>     
>>> the reduced performance/lifetime of hardrives with write cache 
>>> disabled since
>>>       
>> Asterisk
>>     
>>> is not a very write intensive environment.  Even with lot's of
>>>       
>> voicemail
>>     
>>> going on.
>>>
>>> Any other recommendations/links for increasing the reliability of
>>>       
>> Asterisk
>>     
>>> servers?
>>> _______________________________________________
>>> --Bandwidth and Colocation provided by Easynews.com --
>>>
>>> Asterisk-Users mailing list
>>> To UNSUBSCRIBE or update options visit:
>>>    http://lists.digium.com/mailman/listinfo/asterisk-users