The IOPS race is over

emc-f1-carInfrastructure has always been a tough place to compete in. Unlike applications, databases or middleware, infrastructure components are fairly easy to replace with another make and model, and thus the vendors try to show off their product as better than the one from the competition.

In case of storage subsystems, the important metrics has always been performance related and IOPS (I/O operations per second) in particular.

I remember a period when competitors of our high-end arrays (EMC Symmetrix, these days usually just called EMC VMAX) tried to artificially boost their benchmark numbers by limiting the data access pattern to only a few megabytes per front-end IO port. This caused their array to handle all I/O in the small memory buffer cache of each I/O port – and none of the I/O’s would really be handled by either central cache memory or backend disks. This way they could boost their IOPS numbers much higher than ours. Of course no real world application would ever only store a few megabytes of data so the numbers were pure bogus – but marketing wise it was an interesting move to say the least.

With the introduction of the first Sun based Exadata (the Exadata V2) late 2009, Oracle also jumped on the IOPS race and claimed a staggering one million IOPS. Awesome! So the gold standard was now 1 million IOPS, and the other players had to play along with the “mine’s bigger than yours” vendor contest.
Read more of this post

Baking a cake: trading CPU for IO?

Sometimes I hear people claim that by using faster storage, you can save on database licenses. True or false?

The idea is that many database servers are suffering from IO wait – which actually means that the processors are waiting for data to be transferred to or from storage – and in the meantime, no useful work can be done. Given the expensive licenses that are needed for running commercial database software, usually licensed per CPU core, this then leads to loss of efficiency.

Let’s see if we can visualise the problem here with a common world example – Baking a cake.
 
 

Read more of this post

Silly Little Oracle Benchmark – RPM edition

slob-rpmA while ago Kevin Closson announced a new release of the well-known SLOB kit.

SLOB is a simple but powerful toolkit that drives lots and lots of IO on a real Oracle database (so for performance testing of database platforms, it’s much better than synthetic IO tests).

A previous version was bundled with Outrun but required the entire Outrun distribution to work properly. With the new 2.3 version I created an RPM package that can be installed separate on any Enterprise Linux 6.x (64 bit) server.

The wiki page (including instructions) can be found here: SLOB RPM Package wiki

Thanks to Kevin for granting permission to redistribute this awesome toolkit!

Read more of this post

Introducing Outrun for Oracle

Overview

outrun-logo-transparentIf you want to get your hands dirty with Oracle database, the first thing you have to do is build a system that actually runs Oracle database. Unless you have done that several times before, chances are that this will take considerable time spent on trial-and-error, several reinstalls, fixing install problems and dependencies and so on. The time it takes for someone who is reasonably experienced on Linux, but has no prior Oracle knowledge, would probably range from a full working day (8 hours, best case) to many days. I also have witnessed people actually giving up.

Even for experienced users, doing the whole process manually over and over again is very time consuming, and deploying five or more systems by hand is a guarantee that each one of them is slightly different – and thus a candidate for subtle problems that happen on one but not the others. Virtualization and consolidation is all about consistency and making many components as if they were only one.

There are literally dozens of web pages (such as blog posts) that contain detailed instructions on how to set up Oracle on a certain platform. Some examples:

The Gruff DBA – Oracle 12cR1 12.1.0.1 2-node RAC on CentOS 6.4 on VMware Workstation 9 – Introduction
Pythian – How to Install Oracle 12c RAC: A Step-by-Step Guide
Martin Bach – Installing Oracle 12.1.0.2 RAC on Oracle Linux 7-part 1

Even if you follow the guidelines in such articles, you are likely to run into problems due to running a different OS, different Oracle version, network problems, and so on. Not to mention that in many cases the “best practices” provided by various vendors are often not honoured because they tend to be overlooked due to information overload…

Some people have hinted to use automated deployment tools such as Ansible (i.e. Frits Hoogland – Using Ansible for executing Oracle DBA tasks) but there are (as far as I know) no complete out-of-the-box solutions.

EMC has published several white papers and reference architectures with instructions on how to set up Oracle to run best on EMC. Still, some of the papers are not a step-by-step manual so you have to extract configuration details manually from various (sometimes conflicting) sources and convert them in configuration file entries, commands, etc.

So I decided a while ago to go for a different approach, and build a virtual appliance that does all of these things for you while still offering (limited) flexibility in different platform and versions, and preferences for configuration.

Read more of this post

Oracle ASM vs ZFS on VNX

swiss-cheeseIn my last post on ZFS I shared results of a lab test where ZFS was configured on Solaris x86 and using XtremIO storage. A strange combination maybe but this is what a specific customer asked for.

Another customer requested a similar test with ZFS versus ASM but on Solaris/SPARC and on EMC VNX. Also very interesting as on VNX we’re using spinning disk (not all-flash) so the effects of fragmentation over time should be much more visible.

So with support of the local administrators, I performed a similar test as the one before: start on ASM and get baseline random and sequential performance numbers, then move the tablespace (copy) to ZFS so you start off with as little fragmentation as possible. Then run random read/write followed by sequential read, multiple times and see how the I/O behaves.
Read more of this post

Oracle ASM vs ZFS on XtremIO

zfs-asm-plateBackground

In my previous post on ZFS I showed how ZFS causes fragmentation for Oracle database files. At the end I promised (sort of) to also come back on topic around how this affects database performance. In the meantime I have been busy with many other things, but ZFS issues still sneak up on me frequently. Eventually, I was forced to take another look at this because of two separate customers asking for ZFS comparisons agaisnt ASM at the same time.

The account team for one of the two customers asked if I could perform some testing on their lab environment to show the performance difference between Oracle on ASM and on ZFS. As things happen in this business, things were already rolling before I could influence the prerequisites and the suggested test method. Promises were already made to the customer and I was asked to produce results yesterday.

Without knowledge on the lab environment, customer requirements or even details on the test environment they had set up. Typical day at the office.

In addition to that, ZFS requires a supported host OS – so Linux is out of the question (the status on kernel ZFS for Linux is still a bit unclear and certainly it would not be supported with Oracle). I had been using FreeBSD in my post on fragmentation – because that was my platform of choice at that point (my Solaris skills are, at best, rusty). Of course Oracle on FreeBSD is a no-go so back then, I used NFS to run the database on Linux and ZFS on BSD. Which implicitly solves some of the potential issues whilst creating some new ones, but alas.

Solaris x86

slob-rules-kenteken
This time the idea was to run Oracle on Solaris (x86) that had both ZFS and ASM configured. How to perform a reasonable comparison that also shows the different behavior was unclear and when asking that question to the account team, the conference call line stayed surprisingly silent. All that they indicated up front is that the test tool on Oracle should be SLOB.

Read more of this post

Fun with Linux UDEV and ASM: Using UDEV to create ASM disk volumes

floppy-disksBecause of the many discussions and confusion around the topic of partitioning, disk alignment and it’s brother issue, ASM disk management, hereby an explanation on how to use UDEV, and as an extra, I present a tool that manages some of this stuff for you.

The questions could be summarized as follows:

  • When do we have issues with disk alignment and why?
  • What methods are available to set alignment correctly and to verify?
  • Should we use ASMlib or are there alternatives? If so, which ones and how to manage those?

I’ve written 2 blogposts on the matter of alignment so I am not going to repeat myself on the details. The only thing you need to remember is that classic “MS-DOS” disk partitioning, by default, starts the first partition on the disk at the wrong offset (wrong in terms of optimal performance). The old partitioning scheme was invented when physical spinning rust was formatted with 63 sectors of 512 bytes per disk track each. Because you need some header information for boot block and partition table, the smart guys back then thought it was a good idea to start the first block of the first data partition on track 1 (instead of track 0). These days we have completely different physical disk geometries (and sometimes even different sector sizes, another interesting topic) but we still have the legacy of the old days.

If you’re not using an Intel X86_64 based operating system then chances are you have no alignment issues at all (the only exception I know is Solaris if you use “fdisk”, similar problem). If you use newer partition methods (GPT) then the issue is gone (but many BIOSes, boot methods and other tools cannot handle GPT). As MSDOS partitioning is limited to 2 TiB (http://en.wikipedia.org/wiki/Master_boot_record) it will probably be a thing of the past in a few years but for now we have to deal with it.

Wrong alignment causes some reads and writes to be broken in 2 pieces causing extra IOPS. I don’t have hard numbers but a long time ago I was told it could be an overhead of up to 20%. So we need to get rid of it.

ASM storage configuration

ASM does not use OS file systems or volume managers but has its own way of managing volumes and files. It “eats” block devices and these block devices need to be read/write for the user/group that runs the ASM instance, as well as the user/group that runs Oracle database processes (a public secret is that ASM is out-of-band and databases write directly to ASM data chunks). ASM does not care what the name or device numbers are of a block device, neither does it care whether it is a full disk, a partition, or some other type of device as long as it behaves as a block device under Linux (and probably other UNIX flavors). It does not need partition tables at all but writes its own disk signatures to the volumes it gets.

[ Warning: Lengthy technical content, Rated T, parental advisory required ]

Read more of this post

Getting the Best Oracle performance on XtremIO

XtremIO+Stack+NB+copy
(Blog repost from Virtual Storage Zone – Thanks to @cincystorage)

UPDATE: I’ll say it again because there seems to be some confusion: THIS IS A REPOST!

Original content is from the Virtual Storage Zone blog (not mine). Just reposted here because it’s interesting and related to Oracle, performance and EMC storage. Enjoy…

XtremIO is EMC’s all-flash scale out storage array designed to delivery the full performance of flash. The array is designed for 4k random I/O, low latency, inline data reduction, and even distribution of data blocks.  This even distribution of data blocks leads to maximum performance and minimal flash wear.  You can find all sorts of information on the architecture of the array, but I haven’t seen much talking about archive maximum performance from an Oracle database on XtremIO.

The nature of XtremIO ensures that’s any Oracle workload (OLTP, DSS, or Hybrid) will have high performance and low latency, however we can maximize performance with some configuration options.  Most of what I’ll be talking about is around RAC and ASM on Redhat Linux 6.x in a Fiber Channel Storage Area Network.

Read the full blogpost here.

 

The public transport company needs new buses

Future-British-Bus-1A public transport company in a city called Galactic City, needs to replace its aging city buses with new ones. It asks three bus vendors what they have to offer and if they can do a live test to see if their claims about performance and efficiency holds up.

The transport company uses the city buses to move people between different locations in the city. The average trip distance is about 2 km. The vendors all prepare their buses for the test. The buses are the latest and greatest, with the most efficient and powerful engines and state of the art technology.

Read more of this post

Getting the most out of your server resources

hearseespeak

As an advocate on database virtualization, I often challenge customers to consider if they are using their resources in an optimal way.

And so I usually claim, often in front of a skeptical audience, that physically deployed servers hardly ever reach an average utilization of more than 20 per cent (thereby wasting over 80% of the expensive database licenses, maintenance and options).

Magic is really only the utilization of the entire spectrum of the senses. Humans have cut themselves off from their senses. Now they see only a tiny portion of the visible spectrum, hear only the loudest of sounds, their sense of smell is shockingly poor and they can only distinguish the sweetest and sourest of tastes.

– Michael Scott, The Alchemyst

About one in three times, someone in the audience objects and says that they achieve much better utilization than my stake-in-the-ground 20 percent number, and so use it as a reason (valid or not) for not having to virtualize their databases, for example, with VMware.

Read more of this post