How often in your career do you have a chance to dismantle the old architecture in its entirety and build it again - completely redesigned? Usually, a company upgrades the hardware once in a three-four years cycle, while the "backbone" architecture outlives the hardware upgrades and lives dozens of year intact.
I have been "lucky" to work for a company long enough to rethink its backbone DBMS architecture twice: once about ten years ago, when I transformed an old asymmetric architecture into an elegant Warm Standby topology (SAP Sybase Replication Server). This later grew into a heterogeneous replication topology spanning 7 DBMS systems, some ASE others MSSQL, replicating transactions seamlessly across distant cities. Now this is the second time I will have a chance to transform the "backbone" of the site: from active-passive WS topology to an active-active MSA architecture.
Once WS was a thing to contemplate with satisfaction: simple to setup. Simple to maintain (although I did have to build a couple of tools to make the administration more failure-proof). It has proven itself to be vital and stable - close to 10 years of stability - don't remember any failures that resulted from architecture shortages. It was the most simple and elegant alternative for a site that loos for a solution with two available ASEs which may be exchanged without much hassle (I discount storage replication here, which too has its advantages - although its architecture today bears similarity to DBMS transactional replication). Today, warm standby is bulky, constrained, somewhat out-dated alternative, suffering from SPF which needs to be taken care of by out-of-the-box solutions. Especially when compared to an alternatives available around - from the same vendor.
As opposed to it, today MSA replication is considered a "piece of cake." It is easy to setup. Easy to maintain. It gives you most of the things WS has given you once ago, and with much simpler setup. More than that, it is possible to integrate warm standby as a single logical unit within MSA topology - for those in love the the logical connection concept.
In the coming weeks I will be privileged to witness a live transformation of the old good warm standby topology into a new MSA-based topology. Good by my old good friend - I loved to work with you for years. Loved the stability of it. Loved the advantages it gave once. Welcome my new acquaintance. I hope I will live to love you even more.
It has been quite a challenge to think through the architectural redesign. How do you move the heterogeneous replication with 7 sites, some MSSQLs, to an MSA solution with minimal downtime (seconds alone)? As of replication server documentation, the same ASE instance cannot participate in different replication server domains. Not to say that different replication server domains cannot interact or exchange transactions. A tale to be told. But this is over now. This coming weekend the ground will begin to move: the old good pair of M5 servers hosting a set of 15.7 ASEs with a logical axis of 24/7 DBMS instances that distributed data down the stream will start to merge into a brand new pair ot T54 servers hosting a set of 15.7 ASEs united in an MSA replication topology, including a pair of IMDB replicas.
I count myself to be really lucky to be able to push the architecture once again to the forefront of DBMS (and HW) technology. In-memory computing today is something that begins to knock on every door (though few still trust the technology or see its full potentials). Active-active topology too today is something the customers require more and more consistently. Every company today requires its DBMS systems to be available 24/7, to run queries super fast, and to require little to no DBA intervention to "fail-over." IMDB allows you to do things you have stopped to dream about long ago with a ridiculous ease (I don't know of any other DBMS solution that has been as seamlessly integrated into a DBMS product as IMDB - big thumbs to SAP/Sybase product engineers - lovely solution). I have been able to relieve customer's pains strained to meet deadlines waiting for nasty reporting jobs running historically over hours and days with the help of IMDB. We are talking about cutting time from over 48 to 2 hours and less. IMDB is fast - too fast sometimes. Close to zero downtime and close to zero SPF solution is catered by MSA (with all the advantages and disadvantages of transactional replication).
To cut things short - if you have the guts and the vision it takes about 20 minutes and a set of simple, non-intrusive steps to upgrade your whole architecture from warm standby to an active-active MSA topology... If you wont dream - you wont learn to fly.
Pff.... 20 minutes in a lab....
In the real life with TBs of data and site scattered miles away it will take as many hours...
Days?....
Time will tell.
ATM.
- SAP Managed Tags:
