Google Cloud Platform Blog
Benchmarking Web search latencies
Tuesday, March 31, 2015
A few weeks ago we
announced
Perfkit to make it easy for you to benchmark popular workloads on the cloud. As we mentioned, it’s a living benchmark, and we are evolving it to include a new tool to measure the impact on latency when you grow the number of servers that power your application.
We call the new performance benchmark Online Data Intensive Simulator, or
OLDISIM
, written in collaboration with the
Multiscale Architecture and Systems Team (MAST)
at Stanford. It models the distributed, fan-out nature of many modern applications with tight tail latency requirements, such as Google Search and some NoSQL database applications.
We use OLDSIM internally to measure the impact of both hardware and software improvements on our scale out workloads and analyze their scaling efficiency. Scale out efficiency allows us to meet new user demand by adding the fewest number of servers possible while maintaining great user experience. The fewer servers we add, the more energy efficient we are, and the cheaper the solution is. Predicting how a service will scale out is usually very hard under laboratory conditions, but experiments show that OLDISIM results strongly correlate with our current Google Search performance in scaling efficiency, as the chart below demonstrates.
Our needs within Google are similar in many ways to other scale out Internet workloads, and we're making a version of OLDISIM available to the open source community through
PerfKit Benchmarker
. We
shared it
using the Apache V2 license. With OLDISIM, you can more easily model and simulate most applications with a fan-out/synthesis model, including Hadoop and several NoSQL products. You can specify which workload you plug in to each leaf node, and measure the scaling efficiency and tail latency of your applications.
You can run
OLDISIM
by itself by following the instructions on GitHub, or use
PerfKit Benchmarke
r to run it on many of the most popular cloud providers. The command line is as simple as “pkb.py --benchmarks=oldisim”.
Both OLDISIM and PerfKit Benchmarker teams get your feedback through GitHub. We’d love to hear what you think, so please send us your suggestions and issue reports.
Happy Benchmarking!
Posted by Ivan Santa Maria Filho on behalf of the Cloud and
Platforms Performance Teams
No comments :
Post a Comment
Don't Miss Next '17
Use promo code NEXT1720 to save $300 off general admission
REGISTER NOW
Free Trial
GCP Blogs
Big Data & Machine Learning
Kubernetes
GCP Japan Blog
Labels
Announcements
56
Big Data & Machine Learning
91
Compute
156
Containers & Kubernetes
36
CRE
7
Customers
90
Developer Tools & Insights
80
Events
34
Infrastructure
24
Management Tools
39
Networking
18
Open Source
105
Partners
63
Pricing
24
Security & Identity
23
Solutions
16
Stackdriver
19
Storage & Databases
111
Weekly Roundups
16
Archive
2017
Feb
Jan
2016
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
2015
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
2014
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
2013
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
2012
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
2011
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
2010
Dec
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
2009
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Mar
Feb
Jan
2008
Dec
Nov
Oct
Sep
Aug
Jul
Jun
May
Apr
Feed
Subscribe by email
Technical questions? Check us out on
Stack Overflow
.
Subscribe to
our monthly newsletter
.
Google
on
Follow @googlecloud
Follow
Follow
Follow
No comments :
Post a Comment