Backup server SATA vs SAS

[irsprint84]

I m looking at getting a new backup server, obviously getting sata would be cheaper and get more space than sas, would it cause a major issue for a backup server?

[Michael]

Speed is much less of an issue with a backup server, unless you're backing up a ridiculously huge amount of data that it cannot be accomplished over night.

 

Then again, it would be sensible to setup a full backup once a week, then a incremental backup for other days. Either way, SATA will give you considerably more disk space than SAS for a fraction of the price.

[irsprint84]

I currently have a backup server with SAS drives, only 40 - 60 gb mon - friday, does in 3 hours

[mac_shinobi]

you can get RAID Sata hard drives which uses a better ( either controller or firmware or both ) which cost a tiny amount more then a normal sata hdd - how much data are you planning to backup ?

[dhicks]

obviously getting sata would be cheaper and get more space than sas, would it cause a major issue for a backup server?

 

Certainly shouldn't do. I've just re-installed our backup server with OpenSolaris running a ZFS version (22) that supports block-level de-duplication, Samba and a patched rsync 3.0.7 that supports the --copy-devices option.

 

ZFS, using RAID-Z, is implemented by the OS rather than a RAID card. This server has no RAID card, just 6 500GB SATA drives plugged in to the motherboard and a couple of cheap SATA controllers. I get the impression that RAID cards increase disk performance by giving you more bandwidth to each disk, but if they break (which they seem to do more often than you'd think is a good idea) and you've used the RAID card to construct a RAID array from your disks then you're stuck searching around for an identicle RAID controller or you disks will probably be re-initislised when you plug them in to the new controller. A better option might be to use the RAID controller to provide a high-bandwidth channel to each disk but use software RAID (OpenSolaris' ZFS, Linux's mdadm, Windows Server's RAID capabilities) to actually construct the RAID array, that way if the RAID controller fails you can reconstruct the array easily.

 

As ZFS supports block-level deduplication, a server-side file copy operation doesn't actually use up any disk space. Therefore, I can rsync whole directory trees or virtual machine disk images to the backup server every night and only changed data will actually need to be copied. A simple script on the backup server makes a copy of each directory tree or disk image every night after rsync has run. I also make that script copy individual user's files to a backup directory structure shared out by Samba, so any user can go to \\backup001\backup and see a collection of folders, named by date, and inside of each one are copies of their own files from their network user area from that date.

 

The above system is still freshly installed, and is still running initial syncs of large file images and directory trees, but performance certainly seems quite good, and that's just with software RAID, no RAID card.

 

--

David Hicks

[dhicks]

you can get RAID Sata hard drives which uses a better ( either controller or firmware or both )

 

There is also the WD VelociRaptor series of SATA harddrives, which currently seem to go up to 300GB in size, run at 10,000rpm for faster access (standard SATA drives run at 7,200rpm) and might make a good compromise between the speed of SAS and the size of SATA. They are an odd form factor, though, and won't fit standard removeable caddies.

 

--

David Hicks

[glennda]

i have a smart array 60 that is using 6 x 1tb 7.2k sata disks, the weekend backs run until 6am then next morning which is okay as nobody using it, have had no problems with the speed, plus the daily backups are finished within a couple of hours so nobody us using the system while backups are running

[mac_shinobi]

There is also the WD VelociRaptor series of SATA harddrives, which currently seem to go up to 300GB in size, run at 10,000rpm for faster access (standard SATA drives run at 7,200rpm) and might make a good compromise between the speed of SAS and the size of SATA. They are an odd form factor, though, and won't fit standard removeable caddies.

 

--

David Hicks

 

I was thinking that but wasnt sure if the price would come to around the same as SAS - also with ref to the size of the physical drives being smaller you can get a caddy for them - you would have to ask Simon @ cp ltd or search on the different sites ie misco , dabs, cclonline or w/e

[broc]

We use a pair of cheap HP ML110 servers (located in an outlying building at school) stuffed with 500gb SATA drives & flip flop between them for daily DR backups. No RAID, They have drives labelled MONDAY,TUESDAY,WEDNESDAY,THURSDAY,FRIDAY (plus one or two others). The advantage of this 'KISS' approach is you can easily swap a drive if it fails and if the server goes down you can move the drives to anything that will talk SATA.

 

A point to remember, using hardware RAID controllers is fine as long as you have a spare controller available, they are a bit like using tape backup, great as long as you have a working tape drive but what use is a backup tape if the drive is toast?

[Guest Guest]

Certainly shouldn't do. I've just re-installed our backup server with OpenSolaris running a ZFS version (22) that supports block-level de-duplication, Samba and a patched rsync 3.0.7 that supports the --copy-devices option.

 

Please tell...? We've got the same setup here barring the rsync bit, just doing snapshot copies of our VMs atm.

[keogk]

I have been looking into replacing our Tape library with a Disk solution. We already have the Commvault software and Licensing to do it. I was looking at a Case today that held 24 Sata drives and had a 1000W Power Supply. I would love to build 4 of them and fill them with 2TB drives. Just started to look into it. I am going to try and Get 1 of the servers built with 24 500 or 1TB drives to see how well it work. Use Windows storage server but have also been thinking of going with ZFS and Open Solaris. Has anyone done anything with such a large number of drives? I priced out a DIT system today for about 42K US dollars that had 4 4U servers with a Total of 96 2TB WD Sata drives.

[nicholab]

I would go with a raided SATA setup as you should at least be able to get double vs SAS. On disk you what a full backup of at least the week but it could be handy to go back a month. You also want you backup up server to be able to hold bare metal restore point for all servers.

 

I still think tape has a place for medium-long term storage as it does not use any energy and is great for transporting. Today I was restoring data from a tape of 2years ago and I have also got some tapes form 2005 that I have catalogued for possible restore.

Edited February 16, 2010 by nicholab

[dhicks]

Please tell...? We've got the same setup here barring the rsync bit, just doing snapshot copies of our VMs atm.

 

As originally asked over on ServerFault:

 

Sync LVM snapshots to backup server - Server Fault

 

I simply went to the rsync download page, downloaded the latest version and the "patches" tarball and un-tarred both into the same folder. The --copy-devices patch comes with notes on how to apply the patch, but it's simply a case of typing in what's written and pressing enter. Then you recompile rsync and away you go.

 

I've actually spent the last two days sorting out the latest issue, which was that I had to re-install our entire backup server from scratch as the 10GB system drive I was using for it was too small to install GCC. Note that "make install" doesn't seem to do anything much on Solaris - I manually copied the newly-compiled rsync executable over the existing one and everything seems to be working fine.

 

--

David Hicks

[dhicks]

On disk you what a full backup of at least the week but it could be handy to go back a month.

 

I'm waiting to find out how good block-level de-duplication is - before I wiped it over and installed Solaris our backup server ran Linux with my own backup utility on it which did file-level de-duplication. That managed to hold live backups for the past two years, so I'm expecting better performance from block-level de-duplication.

 

--

David Hicks