Monday
Jan282008
DR/BC for web/DB servers
Monday, January 28, 2008 at 4:35AM
All,
I'm looking for a faster/reliable solution for DR/BC as well as for sclability for my web/db servers. I came across VMWare Infrastructure and other products. The I/O performance concerns me to go with virtual servers. I'm also looking into imaging software such as Acrnois.
Could anyone share their thoughts on how it's being done with bigger names such as google/youtube etc..?
Thank you,
Regards,
Janakan Rajendran.
I'm looking for a faster/reliable solution for DR/BC as well as for sclability for my web/db servers. I came across VMWare Infrastructure and other products. The I/O performance concerns me to go with virtual servers. I'm also looking into imaging software such as Acrnois.
Could anyone share their thoughts on how it's being done with bigger names such as google/youtube etc..?
Thank you,
Regards,
Janakan Rajendran.
Reader Comments (10)
Janakan,
What is DR/BC?
Sorry,
Disaster Recovery/ Business Continuity
aha, you want the holy grail :-)
This past thread might also be intereting to you
http://highscalability.com/scalability-vs-performance-vs-availability-vs-reliability-also-scale-vs-scale-out
Really depends what kind of application you are running on the webservers.
I have had good experience with vmware and xen.
You do need a very good NAS for the front-ends and good SANs for the backend.
The rest is all variable.
Thanks Atif,
I know Xen is cheaper than VMware but performance/management wise how do feel about both of them?
I understand about the need of SAN on the backend for DBs. Could you elaborate the need for NAS on the front-end?
Janakan,
Again as I mentioned, most depends on your app. Perhaps you want to elaborate on it a bit.
The NAS on the front-end can be used to share centralized code.
We have around 40 front-end servers sharing a single code-base from one NAS (in high availibility mode).
We use a second NAS cluster for caching (write on server 1, read on server 1-40, delete on server 40, gone from all servers, etc)
You can also use to share the system image from a central servers. Thus removing the need for Acronis.
You have to do it in a way that the NAS does not become the bottle-neck for the systems.
For example:
1. Easy way: system boots off NFS and stay within NFS. (NFS server down, all servers down)
2. More sophisticated way (boot strapping), System boots over NFS, get the script to create the filesytem, get the image to install of the system, get the script to configure itself and woosh it goes doing what its supposed to do. (Systems keep running even if the boot server is down. System got hacked, no-probs, just reboot it, etc)
Hope this helps.
Regarding Xen/Vmware.
I have been happy with both but none of them beats raw power.
Depends on your datacenter design.
You may opt to get 4 highend machines for a 100K and run 20 VMs on them.
Or you can get 40 lowend (I will replace this when it dies) machines which will give you much more performance and will cost you only 50% of the cost. (assuming 40K for the machines and 10K for misc)
Whats your DB and Web Platform ? Maybe someone could suggest some choices.
Most important question to ask and get it from your management is:
Whats the cost of downtime ? How much of downtime and data loss they can tolerate i.e. RTO/RPO ( google them up;) ) and whats their budget.
High Availability does not only include technology, but you need to look into people and processes as well.
HTH.
Atif,
Thanks for your response.
My setup is LAMP and it's more of a content based website (video, images). Upon growth, I'm looking into Webfarm with a load balancer and master-1, slave-N for mysql with a seperate load balancer for slave mysql servers. During the seasonal spikes(after an event) and for scalability I'm looking for a solution where I can take an image and deploy it quickly.
Reg. anonymous post:
Thank you for introducting RTO/RPO terms! If we grow ideally we dont' want any downtime/data loss. So any ideal (I'm sure it'd be expensive!) is ok.
Janakan,
You might want to look at some CDN providers for content mirroring/serving.
You also want to look at Amazon services such as S3.
best regards
Atif
Hmm too many acronyms in this post.
So for the benifit of other readers, this is what I have found
RTO http://en.wikipedia.org/wiki/Recovery_Time_Objective
RPO http://en.wikipedia.org/wiki/Recovery_Point_Objective