Regarding the “No DBA” claim…Vertica’s architecture greatly reduces the time DBAs need to spend getting it running and keeping it fast (via all sorts of automatic installation, tuning, physical design, replication, failover, recovery features as Robin describes). DBAs ARE still required (as LewisC mentions) to focus on data security, backup and integrating new data into the database. Vertica spares DBAs from the design, tuning and recovery tedium so they can focus on higher value added work.

Regarding Niko’s orders of magnitude math…Good catch! There are examples of >2 orders of magnitude performance out there. One customer, NetworkIP (a US-based prepaid telecom services provider) described (in a webinar) a query that formerly took 12 hours to run and now takes 10 seconds…depending on the workload, customers will find these exceptional boosts in query performance (for a given query), but for the overall query workload (averaged across tens of different queries a given DBMS is handling), the query performance increase typically ranges from 50x to 200x relative to a row-oriented DBMS.

Lastly, regarding Adi’s comment about whether a column DBMS performs as fast when processing more than just a few columns…A column database reads just the columns referenced by the query, which reduces disk IO and speeds up query handling (versus an index-less row db, which must scan the whole table). But if the query is referencing every column in the table (a rare query for most data warehouses), this benefit was negated in early column dbms implementations. The Vertica architecture has much more than column-orientation going for it…data compression (typically up to 90%, and Vertica can query data in its compressed form) further reduces disk IO, and as Robin mentioned in his comment, it’s shared-nothing architecture and automatic data replication spreads the query and loading work across multiple servers working in parallel. So even when reading a relatively large % of the table, Vertica performance stays fast.

As LewisC mentioned, the Vertica Architecture white paper (found on the vertica home page) describes how all this works in detail.

This is an interesting thread–thanks all.

Adi

I got there by making a mathematical error. Thanks for the heads up. I’ve now corrected it in the article.

3 hours (10800 seconds) to 10 seconds is 1080 times (3 orders of magnitude) faster.

Provider: Backups, DR planning, tuning, daily maintenance, etc. This stuff all still needs to be done, it just shifts from local to remote.

Customer: Creating users & roles (security), monitoring performance (to decide when to add capacity), schema design, additional indexing, etc.

The cloud changes where things run, day to day still has to get done.

LewisC