Ask the Experts – TechSelect* 2008 Thursday 6 November, CBI Conference Centre Sybase Experts: Steve Hookings, Ian Macleod, Mark Riddoch, Paul Dow, Bart Van Kuijk John Trollope, UNUM Provident Ltd What is Sybase’s official position on virtualised servers for Windows 2003? My company are increasingly moving into server consolidation and particularly trying to move away from physical machines and go to virtualised servers. Paul Dow said he would get back to John with the official answer. He went on to say that Sybase are using virtual servers extensively within Sybase engineering. The typical stance is that if the problem reproduces on a non virtual server then it is a genuine problem. FollowUp: The official statement is located at in item #6 in http://www.sybase.com/detail?id=1032794 Steve Hookings said that one of the problems with Windows is that whenever you install ASE it messes with the registry (which is bad news) and therefore the easiest thing to do is create a virtual server and possibly patch it to a certain level of the OS and a certain level of ASE. We do that a lot in engineering for repros and as engineers we don’t have any problems with them but I wouldn’t suggest you run that in production. Ian Macleod said that as an engineer he develops on a MAC and has about 6/7 different virtual machines running under MAC Fusion – he said he wouldn’t run Windows on hardware. He said he has several different versions of Linux and Solaris X86 etc and said it is a fantastic development environment because of the snapshot ability. E.g. I’ll take a snapshot and do something radical to the operating system and if it doesn’t work I can just back it out and roll back on the transactions. John Bidwell, Citi In Terms of ongoing engineering support of the Linux X86 Platform – is that something you are moving towards? Mark Riddoch: To demonstrate the importance of Linux X86 Mark said that Sybase initially only released cluster on the two most popular platforms which were Solaris and X86 Linux. John Bidwell, Citi Are Sybase seeing a pick up on Linux cases? Mark Riddoch: We see a lot of Linux cases coming in, but I can’t say the cases are growing. Steve Hookings: As an engineer, if I am working with the Optimizer, most of the problems are generic so by the time I see the case I don’t know what platform it came from (although I could find out). John Daly, BGI Re migrating to ASW 15 – How can we replicate production load on a development system and test it. Does Sybase have any tools like QP Tune that Sybase can recommend for this? Basically we want to run production load on the development server and check the performance. Ian Macleod: Since this was brought up a year ago at TechSelect this is been a personal quest and something that has been considered very seriously i.e. to capture workloads and re play them in production. There are some 3rd party solutions which we are looking at and how we could integrate or extend them. One of the directions we are looking at (don’t quote me on this) is aiming towards something like partition/index advisors. We want to be able to have the optimizer go a couple of steps beyond the show missing stats thing and say this is where I want you to put your indexes and partitions in order to increase your performance - part of this would be the replay mechanism because the advisor would need to not just make blind hypothesise it needs to test them by creating “what if” indexes; indexes that don’t really have any real index stuff behind it but sufficient to allow the optimizer to do its job in estimation. i.e. an index advisor might look at the performance of the original play and say an index may be better here, here and here so why don’t we create these “what if” indexes and replay it to see what the optimizer would have done. John McGuire, Nomura The Database dump is as, of when the command ends and not when the command starts, which gives me a problem in that my applications overnight shut down for the duration of the dumps running and then they cannot start up again until they have finished. Can the experts suggest a way I can return to the days of 4.9.2 when you started the dump, that was the image and it was a known point in time. Can I have the dump when I start the command and not some time when it gets round to finish? Also: If I had to restore to a point during the 1 hour that the dump was taking place, how would I do it? Paul Dow said he would ask Graham Ivy to get back with an answer Follup Up: There is no mechanism to revert back to the very old behaviour. The current behaviour is specifically designed to maximise the capturing of changes made during the dump period. Lakhbir Gill, RBS Global Banking Re shrinking databases: We are a multi vendor and can do this on other databases but not on Sybase. Paul Dow: This has definitely been looked at. It has been a hot topic for a long time, there has been lots of interest and a lot of effort spent looking at it – it sounds simple but it is a lot more complicated and very difficult under the covers but I would be optimistic of seeing something in a future release. Will Sybase consider providing a configuration wizard that will come up with a series of recommendations based on the system settings that it can detect from the server? As a team leader I want to be able to say to a junior person with not much database experience that I want them to get Sybase up and running and install the databases. The problem is if you install Sybase out of the box and load the databases non of the memory configuration will have happened and you will therefore have an unusable database. We have taken time to write out memory configuration suggestions including formulas and I am sure that this manual process could be programmed. A Sybase DBA is generally more experienced than on other platforms but it is difficult to recruit experienced Sybase people – Paul Dow: This question has cropped up a few times but I haven’t seen a commitment yet in any product roadmap. However, given that we are trying to make things ever easier and this shouldn’t be too difficult, I wouldn’t be surprised if it doesn’t appear somewhere soon. Follup Up: 15.0.3 ESD#1 has started to introduce this ability when a new server is created. Look for OOB (Out Of the Box) configuration. Paul Davies, Barclays Capital Product Enhancement Request: The thing that causes me the most sleepless nights is re- indexing and data stats jobs going belly up. I would like to see re-indexing and data stats amalgamated and get a re-org command with index stats. I.e. could re-index or re-org do the stats as it is doing the scan so that afterwards there is no need to do an update stats. Steve Hookings: that is a fair point – I would suggest you raise the case. Bart Van Kuijk said he would contact Paul on the issue. John Daly, BGI Are there any plans in ASE 15 that when you do a read that you take no locks but get committed data – i.e. to get a consistent snapshot that won’t get locked in the write. Ian Macleod – this is a big project has been debated internally for a long time; it would be quite a change for us but it is being evaluated. Steve Hookings – this is a question for Graham Ivy What are your plans for ASE 16? Clearly there are plans for the next release, but at this stage we don’t know what or when 16 is going to be. Also, although we have a lot of internal names of releases we can’t say what numbers will be associated with those. Paul Dow – audience poll We are looking at the possibility of getting rid of shared memory. It isn’t decided or something that would be very easy, but to help with our planning I would like to ask; if shared memory disappeared would that be an issues for anyone – or would it be positive? It would look similar to NT. With Shared memory dumps how big an issue is it in the downtime of the server. If you take as CSFD you have a couple of ways of doing it; the safest one which guarantees the best information is how the engine is halted which means that during the CSFD things freeze. There is a lot of memory and it can take a lot of time even if we do things like paralyse the dump. For how many people is the time the CSFD takes an issue? Bart Van Kuijk said that the times this is reported to customer support have come down significantly over the last 2-3 years – before times were up to 20 minutes where as now it hardly ever seems to take longer than 30 seconds. He asked the audience if people had been put off doing shared memory dumps for fear that it would take a long time? Paul Dow - audience poll Regarding shared memory; size of shared memory dumps and the time it takes to get them into Sybase – how many people have problems with that? The problem was registered.
Pages to are hidden for
"Ask the Experts"Please download to view full document