Hello All Gurus,

I would like all of you to help me designing the best possible approach to overcome failure and to achive optimal availability of database server.

I need to particularly think about designing a database layout in which I am thinking of using two datacenters sites for placement of my database servers to overcome the failover in these two particular cases:

1. Any one component at a data center going down (Just 1 database server/machine going down)
2. Full data center going down (catastrophic failure)

So what I am thinking so far is to have two database servers at each site which could provide backup of each other (i.e. total 4 servers) now I need to arrange and use these servers in such a way that could have the optimal backup and fault tolerant environment/setup. The reason why I am using two servers at each side is that I don’t want to route the whole database traffic to remote backup site in case of single DB machine failure. Or In other words this way I way trying to add another layer/cluster of database server. Does the Informix provide any cluster solution which could solve my problem?

I hope now you have understood all my thoughts/ideas. Kindly suggest me the best possible way and mechanism which I could use to get the optimal results and availability of my database servers.

I was reading various HDR documents to get any kind of clue about adding various secondary servers for a single primary, and then implementing the best failover strategy to ensure maximum availability of database server to out applications.

I have also explored Enterprise Replication to get my desired result but it seems more complex, also let me know that if ER could replicate the entire database at the other server just like HDR i.e. I want to replicate all my database objects (Tables, procedures, views, triggers, DDL, DML etc.) to the other database server working in the Enterprise replication environment. If it supports all this then kindly tell me the way how I could use ER in conjunction with HDR to get my desired layout.

Or If you feel there can be other ways/possibilities to achieve the above goal then you are most welcome to suggest that.

I shall be very thankful to you.

Thanking you in advance for your anticipation.

Let me know in case of any further query/explaination.

Regards,
Omer Saeed Khan

Hi Omer,

Just some info to get you on your way (I hope) ....;-)

HDR: Is typically made for these jobs, however is supports only 2 nodes (a primary and a secondary)

ER: Is not really designed for this. It is designed to share data. It can replicate all tables of a databse, but no other database objects and new tables etc. are not automatically replicated. So this might be possible but has soem drawbacks.

Possible solutions I see at this time:
I don't know what you platform is, but some hardware vendors like sun can deliver clustering on filesystem/raw-devices. When setting this up at the two centers and betwwen the centers replicate with HDR is a possibility.

The ER/HDR combination is also possible. I would then on both centers et-up a cluster with HDR and replicate between the clusters with ER. The reason I would choose this order is that ER can better handle "line" outages, and has the ability to compress the data.

The third and maybe best solutions is to wait for IDS Cheetah. There are rumours it will support HDR on multiple nodes (and even shared disks etc.) I've been involved with the closed and open beta, but unfortunately haven't seen this option yet. It might be in a next beta drop. If so it would solve all you problems I think. I don't know if you have the time to wait??

Greetz,

RobP

Hello RobP,

First of All, i would to thank you for your very through response.

I have been doing some R&D at my end and have come up with the same solution that you have proposed me i.e. using ER/HDR combination. I am using HDR between two servers which would reside within same data center and similar for two servers placed at other center. And then i will put the ER (Primary-Target mode) between the 2 primary servers of HDR pair placed and two datacenters.

Now i need to further evaluate that design and would love to have your further input,ideas,suggestions,precautions etc. which might be relevant to this strategy.

Kindly do let me know about your point in which you mentioned that "ER can better handle "line" outages, and has the ability to compress the data. " ??? How to compress the data between data centers and to handle the line/network outages???

I shall be waiting for your prompt response.

Thanking you in advance for your anticipation.

Regards,
Omer Saeed Khan

Hello RobP,

First of All, i would to thank you for your very through response.

I have been doing some R&D at my end and have come up with the same solution that you have proposed me i.e. using ER/HDR combination. I am using HDR between two servers which would reside within same data center and similar for two servers placed at other center. And then i will put the ER (Primary-Target mode) between the 2 primary servers of HDR pair placed and two datacenters.

Now i need to further evaluate that design and would love to have your further input,ideas,suggestions,precautions etc. which might be relevant to this strategy.

Kindly do let me know about your point in which you mentioned that "ER can better handle "line" outages, and has the ability to compress the data. " ??? How to compress the data between data centers and to handle the line/network outages???

I shall be waiting for your prompt response.

Thanking you in advance for your anticipation.

Regards,
Omer Saeed Khan

Hi Omer,

First let me go into the compression. With CDR_NIFCOMPRESS onconfig parameter the compression level can be set. This takes up some CPU time, but saves network bandwidth.

Second issue: the line outage protection
HDR has some outage protection. It can resync as long as the logical log files have not been wriiten over (logical log on disk are somewhat circular). As long as this isn't the case, it can automatically correct the line outage when the link comes up again. But after that the only solution is a backup/restore.

With ER the logsnoopers that are looking at the logical logs will transfer thye data to the other server(s). Whenever the line goes out and the quememmory also runs full, it will send the data that need to be send later to a stable queue on disk. If you make the diskspace large enough it could deal with very long line outages.

Hope this answers your questions,

RobP