Well 11g is officially “launched”. I think that just means you can download the white papers from OTN now. Seems that the actual software itself still isn’t available although there was talk that it would be released by August or something – on Linux at least. I’m not exactly holding my breath although I do think they’ll have it out soon, as promised.
I missed the first hour of the Oracle 11g Launch because I was at the Illinois Secretary of State trying to get license plates for the car I just bought. I took some vacation time last week for a trip to visit my family in Michigan and bought a used car while I was there. (I’ve been looking for a few months now.) Trying to have all the right paperwork for the government is always a hassle and I had to leave twice and upon returning always ended up queuing in line again. While in line the third time, I actually remember standing there thinking about how this was similar to database processes that can’t obtain a latch by spinning and have to sleep. Perhaps my vacation wasn’t long enough.
Anyway… it was interesting to see which new features got top billing. Not many surprises (that I noticed anyway). Real Application Testing seemed to get a lot of enthusiasm. It also has the top spot on the list of new features at the newly available 11g page on OTN and in the press release.
They displayed “benchmarks” for the new SecureFiles… giving the same read performance as a filesystem and better write performance. I’ll want to try that myself, or at least see some more details. One odd thing was that they called it “Oracle Fast Files” throughout the launch and in the launch slides and in the press release. Maybe there was a last minute name change for this feature.
The feature that jumped out at me the most was compression. Andy Mendelsohn talked about it during his presentation on ILM and Yahoo talked about it some more right after that. Compression has existed in past versions of Oracle however it seems to have some significant improvements in 11g. For example, it can be used on transactional workloads with very little overhead.
To date I haven’t seen large numbers of people using compression in Oracle. Up to now the only people who used it were data warehouse operations. However I think that this technology is going to start becoming more and more important. This isn’t so much an Oracle-specific thing; actually I think that it will become more important in the IT industry as a whole.
The main reason is because of where I see hardware going. At Google’s recent Seattle Conference on scaleability, Peter Braam from Cluster File Systems Inc (home of the Lustre project) discussed this during his talk [about 43 min in]. Processing power is increasing as we pack more cores into sockets. Disk capacity is increasing as well. But disk bandwidth is not increasing. Those little round platters can only physically spin so fast – and so it’s not likely that seek times are going to decrease dramatically without a breakthrough in classical mechanics. In fact disk access times will actually get worse – by storing more and more data on larger and larger platters we will increase contention for those heads and I/O requests will likely spend more time waiting in queues. Maybe if SSD got real cheap… but I don’t see that as likely anytime soon either…
There are two solutions to this problem:
Increase aggregate bandwidth by using more spindles.
Write less data.
Increasing spindles is database architecture 101. However there is a practical limit
as spindles add some cost and management overhead. (The limit is lower for smaller organizations.) The next step is compression – and Oracle 11g appears to have made some good strides. It’s really no different then setting the appropriate PCTFREE (and PCTUSED for MSSM tablespaces) – you’re just trying to get optimal space and I/O utilization. Historically there has always been a CPU penalty for using compression – but Oracle appears to have made some progress with this for 11g. Of course there will be some overhead but this could be offset by the advancements in processor speed and the resultant I/O savings.
Compression is not without drawbacks; for example administrators accustomed to manually inspecting blocks will find that this doesn’t work anymore. I’ll have to try out bbed on 11g and see what the differences are for that utility with compression. Also, perhaps most significantly, Oracle’s current pricing model really encourages customers to try to offload as much processing as possible and use configurations with the minimal number of CPUs possible. But notwithstanding the licensing problem I think that data compression is a technology whose time has come and I think that we should start seeing quite a bit more adoption in the coming years.
Maybe sometime I’ll try to write a bit more about this stuff… but in the meantime I’m looking forward to trying out that first 11g release as soon as it’s available!!