Услуга · ИТ-инфраструктура

System performance tuning

When everything starts crawling the first instinct is to buy a more powerful server. Often that does not help: the bottleneck is the disks, the database configuration or a saturated network port, and the new processor sits idle exactly as the old one did.

Measurements
instead of guesswork
Without buying anything
at least to begin with
The numbers
before and after the work
Report
with the real causes

What the work includes

We work from measurements. Until it is clear what the constraint is, any purchase is a lottery.

Discuss the scope

Measurements under live load

We record processor and memory load, disk queue length and traffic precisely while the office is working.

Reviewing the servers

Where the ceiling is: not enough resources, a poor configuration, unnecessary background processes, antivirus scanning the database files.

Databases

Whether the right indexes exist, when maintenance jobs run, and locking conflicts under heavy use.

Network

Port errors, mismatched settings, saturated links between nodes.

Configuration first

Most findings are cured by configuration rather than money.

Measurements afterwards

Repeat measurements so the result is proved in numbers rather than asserted in words.

How it goes

The survey fits into a working week, and the time to fix things depends on the findings and is often measured in hours.

01

We listen to the complaints

What exactly is slow, for whom and at what time. Specifics are more useful than «everything is slow».

02

We measure

We watch during the busiest hours and record the metrics.

03

We fix

We start with the free things: settings, exclusions, moving scheduled jobs to the night.

04

We reconcile

Repeat measurements and comparison with what was.

Antivirus without exclusions on a database server is a finding that comes up constantly. Scanning the database files on the fly slows everything by several times. Adding the exclusions the vendor recommends is half an hour's work, yet people complain about the slowness for months.

Questions and answers

Sometimes that is true, but you can only be sure by measuring. If the bottleneck is the disk subsystem, a faster processor changes nothing. A survey costs a small fraction of the hardware and regularly proves that no spending is needed at all.

In a typical case tuning gives about a one-and-a-half-fold speed-up. More than that comes where a crude mistake turns up: a missing index or a heavy scheduled job that for some reason runs in the middle of the working day.

By numbers rather than by feel. The metrics are taken before the work and again afterwards, and the difference is set out in figures. Subjective impressions change with the mood.

We will find out what exactly is slowing things

Describe which operations drag and for which staff. We will take the measurements and name the real cause.