the math behind availability
DESCRIPTION
a presentation about the mathematics behind availability of IT servicesTRANSCRIPT
dd-mm-yyyy
© For internal use
dd-mm-yyyy
© For internal use
Availabilitythe math behind…
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Definitions
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Basic maths ?
𝐴=𝑀𝑇𝐵𝐹
(𝑀𝑇𝐵𝐹+𝑀𝑇𝑇𝑅)
𝐴=𝑈𝑝𝑡𝑖𝑚𝑒
𝑈𝑝𝑡𝑖𝑚𝑒+𝐷𝑜𝑤𝑛𝑡𝑖𝑚𝑒
Simplest Availability formula
More “design” formula
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Availability 101
“The system will be up if both Nodes 1 AND 2 are NOT down.”
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
The math’s : Availability in Parallel
𝐴=1−(1−𝐴𝑥)2
Ax
Ax
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
The math’s : Availability in Series
𝐴=𝐴𝑥 𝐴𝑦
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Availability Computation…
MTBF ?
MTTR ?
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Estimating Availability
Component Availability MTTR
WAN Link 0,998 8 HRSRouters/Firewall 0,999 4 HRSSwitches 0,9999 4 HRSServer 0,999 4 HRSDC Tiers3+ 0,99982 Years ?
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Let’s apply all those things …
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
2 nodes cluster…
Active/Backup Active/Active
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
2 nodes cluster : Failure ?
Both nodes fails
One node fails, and the system is in the process of failing over
One node fails, and the failover fails
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Two node fails…
p(dual node)
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
2 nodes cluster : “Cost of Failover”
Cost of Failover
0.999 0.99 0.99999
MTFO
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
“Failover” …
𝑝 ( 𝑓𝑎𝑖𝑙𝑜𝑣𝑒𝑟 )= 𝑚𝑡𝑓𝑜𝑚𝑡𝑏𝑓 =(1−𝑎)
𝑚𝑡𝑓𝑜𝑚𝑡𝑟
mtfo
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Failover Faults !
𝑝 ( 𝑓𝑎𝑖𝑙𝑜𝑣𝑒𝑟 𝑓𝑎𝑢𝑙𝑡 )=(1−𝑎)×𝑑
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Active/backup vs Active/Active
𝑝 (𝑠𝑦𝑠𝑡𝑒𝑚𝑑𝑜𝑤𝑛)=(1−𝑎)2+ (1−𝑎)× 𝑚𝑡𝑓𝑜𝑚𝑡𝑏𝑓 + (1−𝑎)×𝑑
0,999968 0,999998
Active/Backup Active/Active
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Next ? 5 nodes vmware cluster…
d
Single Host ~99,95 %
4 Hosts/ 1 Spare ~ 99,9984 %
4 Hosts/ 2 Spare ~ 99,9986 %
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
An Application with DRS
DC1 DC2
99.99
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Availabilty of Each DCs…
0,999999969 0,999999969
DC1 DC2Mtfo = 20 min
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Availability of Both DCs…
0,99999999 0,999999998
DC1 DC2
0,999999998Final 0,999999977
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Lesson 1 : K.I.S.S.
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Human factor
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Lessons 2 : Redundancy
03-05-2014 | Valery Brasseur | © For internal use TO | FDS2 | MSHD
Lessons 3 : Good Failure Detection
Worldline is a registered trademark of Atos Worldline SAS. June 2013© 2013 Atos. Confidential information owned by Atos Worldline, to be used by the recipient only. This document, or any part of it, may not be reproduced, copied, circulated and/or distributed nor quoted without prior written approval from Atos Worldline.
03-05-2014
© For internal use
Worldline is a registered trademark of Atos Worldline SAS. June 2013© 2013 Atos. Confidential information owned by Atos Worldline, to be used by the recipient only. This document, or any part of it, may not be reproduced, copied, circulated and/or distributed nor quoted without prior written approval from Atos Worldline.
© For internal use
ThanksFor more information please contact:T+ 33 3 20607982M+ 33 6 [email protected]