May 26, 2010 at 3:39 pm
Hiya
I want to maintain a DR solution at AWS - I plan on having three VMs: MSSQL, Primary App/Terminal Server, *nix file server. My plan is to ship my full, diff, and tlog backups continuously to the unix file server on AWS while keeping the other two VMs asleep. Regularly I will boot up the other two VMs, patch them and slurp up the latest backups so the DB is close to current. Note that our production/live DB can be read-only at night, we do updates during the day only and it is okay for it to take several hours to spin up in case of DR [DR in my world is a natural disaster threat preventing staff from entering the office for safety reasons, or otherwise the building disappearing].
In case of a DR event I'll boot up the DR MSSS box and apply full/diff/tlog and then turn on the app/terminal serving VM. Primary DB storage for the VM DB is a RAID5 array of virtual disks.
Where am I going to trip up? Any suggestions / concerns?
thanks!
June 1, 2010 at 5:18 pm
is this really not an interesting topic?
Doh!
June 1, 2010 at 5:31 pm
I'm not sure what you mean. Are you saying that you'll have the VM files on Amazon's service or you'll host VMs at Amazon? I'm not familiar enough with the Amazon offerings or performance here.
Likely the issues you face are the bandwidth to make the transfers, the cost of Amazon's service, and then how well it will perform in a DR situation.
Is Amazon better than some box at Rackspace or another vendor? No idea. You'd have to test, or experiment. In theory Amazon is no different from using Rackspace, but I don't know how it will work.
June 1, 2010 at 5:54 pm
Hi Steve!!
Great work - I'm a daily reader here!
Anyway, apologies if my jargon is off 🙂
Are you saying that you'll have the VM files on Amazon's service or you'll host VMs at Amazon?
I meant that I will run three 'machines' in their cloud - those guys basically run a version of Xen to host virtual machines that are chosen (or even modified) by their customers (me!). By VMs I meant virtual cloud based computers (guest OSes)... which in Amazon's case are Xen hosted virtual machines (Xen managed by Amazon, VMs are 'mine').
Likely the issues you face are the bandwidth to make the transfers,
My DB isn't that big: I produce about 35GB/month in transaction log backup files and about 15GB/month in DB backup files. The TLOGs compress down about 10:1 and the backups about 4:1. I do TLOG backups frequently - it comes out to about 5MB/hour which is no biggy latency and capacity-wise. I'm doing full backups on the main DB weekly off-hours [1GB, compressed, on Sunday] and the other DBs nightly [essentially zero size]. Having said that I'll add that our rate of growth is increasing rapidly as the company grows - for good reason, we are a young company and just hit the 'steep part of the hockey stick' late last year; thus I anticipate our transfer volume to go up substantially but we are still far from having bandwidth or even latency issues with our Internet links.
the cost of Amazon's service,
This is the best part; based on my own sizing estimates, rooted in our current production environment and load, I've compared against a few hosting options including leasing machines at a robust hosting site and leasing rackspace in same ... there is no comparison. As far as I can tell the cloud is cheaper by far for my needs. I am looking at Amazon for no great reason compared to Rackspace - I use Rackspace at a different company and I like them a lot.
and then how well it will perform in a DR situation.
I'm not sure here. I assuming if I go with a beefy virtualWindows Server 2008 64 bit box and use enough striped EC2 drives I should get plenty of I/O and CPU out of it. The remaining aspects then are the primary application (entirely DB bound), windows terminal server for remote access during a DR event (ie remote working until we are back in our office - or a new office), and of course network bandwidth for our terminal server users.
June 1, 2010 at 9:37 pm
It sounds like you've thought it through. Assuming Amazon gives you good bandwidth and enough horsepower for the VMs, you ought to be fine.
The transfer is the only thing I'd be careful of. Isn't Amazon starting to charge a fee/GB this month for transfers in?
June 1, 2010 at 11:11 pm
The transfer is the only thing I'd be careful of.
In-bound bandwidth will be 10c per GB + they charge 10c per million I/O requests (max 4KB@).
If I include weekly patch + DB Load, quarterly DR testing for a day each, a 5 disk RAID array for the live DB files plus the OS, a 100% mirrored clone of all filesystems (call me paranoid) throw in CPU, windows licensing, larger size virtual iron (for the RAM and IO capacity) for both the Windows DB box and the *nix file server, it's still under $1200/year. Actual DR days will cost about $75@.
Looks to me like leased rack or hardware at a reputable facility cannot compare. Once I've tried this more than zero times in a row I'll feel better about it, I wanted to make sure I didn't miss anything obvious before getting too far down the path.
June 2, 2010 at 8:50 am
That sounds like a good plan then. I'd give it a shot, short term, test the DR, and post back. Or write up your experiences. I'm sure people would be interested in how it goes.
June 2, 2010 at 8:59 am
Will do.
cheers
ram
Viewing 8 posts - 1 through 7 (of 7 total)
You must be logged in to reply to this topic. Login to reply