Operations in a Distributed World - The Practice of Cloud System Administration - page 147

means SRE efforts are focused on a few select business metrics rather than being pulled in
many directions by users, each of whom has his or her own priorities.
Another difference is in the attitude toward uptime. SREs maintain services that have
demanding, 24 × 7 uptime requirements. This creates a focus on preventing problems
rather than reacting to outages, and on performing complex but non-intrusive maintenance
procedures. IT tends to be granted flexibility with respect to scheduling downtime and has
SLAs that focus on how quickly service can be restored in the event of an outage. In the
SREview,downtimeissomethingtobeavoidedandserviceshouldnotstopwhileservices
are undergoing maintenance.
SREstendtomanageservicesthatareconstantlychangingduetonewsoftwarereleases
and additions to capacity. IT tends to run services that are upgraded rarely. Often IT ser-
vices are built by external contractors who go away once the system is stable.
SREsmaintainsystemsthatareconstantlybeingscaledtohandlemoretrafficandlarger
workloads. Latency, or how fast a particular request takes to process, is managed as well
asoverall throughput.Efficiency becomes aconcernbecause alittle waste permachine be-
comes a big waste when there are hundreds or thousands of machines. In IT, systems are
oftenbuiltforenvironmentsthatexpectamodestincreaseinworkloadperyear.Inthiscase
a workable strategy is to build the system large enough to handle the projected workload
for the next few years, when the system is expected to be replaced.
As a result of these requirements, systems in SRE tend to be bespoke systems, built on
platformsthatarehome-grownorintegratedfromopensourceorotherthird-partycompon-
ents. They are not “off the shelf” or turn key systems. They are actively managed, while IT
systems may be unchanged from their initial delivery state. Because of these differences,
distributed computing services are best managed by a separate team, with separate man-
agement, with bespoke operational and management practices.
While there are many such differences, recently IT departments have begun to see a de-
mand for uptime and scalability similar to that seen in SRE environments. Therefore the
management techniquesfromdistributedcomputingarerapidlybeingadoptedintheenter-
prise.