Re: IDE DAT Drive?
On 2000-11-23 22:19, Robert Davies wrote:
>On 2000-11-23 09:20, Robert Davies wrote:
>> Do you REALLY have 300G of things that need to be backed up?
>> If so the solution is a DLT robot. But most people who think that they
>>have that much data to backup aren't doing things effectively.
>I sure do, I wish I hadn't and it's growing 1GB a week. Unfortunately the
>archives get updated,
>so I can't even freeze parts of it, it all has to be online 24/7.
That's going to be tough to solve. If you backup 300G over 100baseT
networking (typical network backup situation) then it'll take at least 9
hours if you have an Ethernet segment dedicated to it and the data is all in
large files. This is probably unmanagable so you'll need a tape drive in the
server (which can cause other problems). I recommend not having tape drives
in the server if you can avoid it as I have seen too many tape drives lock up
SCSI busses and bring servers down.
>> Don't backup news spools!!!
>> Don't backup email as it takes too long to backup and changes too fast.
>Add squid cache's and the like to that.
>>If you have a serious failure then users will be upset regardless. If you
>>use a mail server such as Netscape's mail server then due to it's databases
>>you would have to restore while the server is offline. Recently I was
>>working for a large ISP (>750K users) and I estimated that it would take 10
>>days to complete a backup of email. If email takes 10 days to restore then
>>you are better to dump it. If it takes 5 days to restore then it's probably
>>best dumped. How fast should a restore be to be worth-while?
>In that case yes, but I find it scary that they have all their eggs in one
>basket. Prefer a divide and conquer approach, have enough problems with
>Oracle DBs where it's all or nothing, most times.
Well that's the way that Netscape works. They do lots of other interesting
things too such as delayed deletion of messages which seem designed to work
around performance problems in the Sun UFS file system (run it on ReiserFS or
even Ext2 and you don't need all this junk).
>>I had a big discussion on this topic with my former colleagues. They
>>couldn't understand why backing up a mail spool (tens of millions of small
>>files) would be slower than backing up a web server (tens of thousands of
>>mall files, a few really big log files, and the OS).
>If you've that many files in a mail spool then you have problems.
Not necessarily. You only have problems when you run out of disk space and
it's almost impossible to enlarge the space because you can't back it up and
the hardware RAID array decides not to allow you to enlarge the volumes.
>> low quality versions of the parts for IDE. Also compared to the cost of
>> tapes the cost of a SCSI controller is nothing.
>Yep, though Matt could comfort himself with the though that the Onstream IDE
>has been used successfully and discussed on the Linux kernel mailing list.
Sure. But you have to ask "how much will it cost in my time to fix things
and how much in my colleagues' time if there is down-time" and then you
usually come to the conclusion that buying a more expensive device that will
"just work" makes economic sense.
I am all for IDE hard drives and CD-ROM drives. But all the evidence I have
seen shows that SCSI is better for tape drives, CD-R, CD-RW, etc.
http://www.coker.com.au/bonnie++/ Bonnie++ hard drive benchmark
http://www.coker.com.au/postal/ Postal SMTP/POP benchmark
http://www.coker.com.au/projects.html Projects I am working on
http://www.coker.com.au/~russell/ My home page