Upload
silvia-horn
View
215
Download
0
Embed Size (px)
Citation preview
STORAGE ARCHITECTURE/MASTER:
SAN Math for Core/Edge SANs
Spicing it Right!
Norman OwensIndependent Storage Consultant
SAN Math for Core/Edge SANsSpicing it Right!
Preview
• Distinctions between topologies
• 5 critical variables for sizing:
S.P.I.C.E.
• Comparative sizing
Distinctions among topologies:3 topology types
Starting point: Island(s) of SAN
A scaling design: Collocated SAN or “CoLo SAN”
A scaling design: Core/Edge SAN
A SAN entry point
Island(s) of SAN
A starting point. Buy
edge switches and disk
as needed.
A scaling design: CoLo SAN
A CoLo SANCluster servers and their
storage in functional units
on edge switches. Same as
Islands but with Director
added for the “any-to-any”
connections.
A scaling design: Core/Edge SAN
A Core/Edge SANPlace storage and critical servers on Director class switches and put all regular servers on edge switches.
0/0
20% 20% 20% 20% 20%
1 2 3 4 5
Topology types: Which best describes your SAN plans?
1. Mostly Islands of SANs2. Moving to Core/Edge with
some Islands remaining3. Moving to CoLo with some
Islands remaining4. Plan to link islands with tools
outside of simple fibre channel connectivity
5. Meshing Directors together with few edge switches
0/0
20% 20% 20% 20% 20%
1 2 3 4 5
Why do you have Islands?1. It just happened
2. Department/Organizational
structure encourages it
3. Islands bring stability by
limiting scope of SAN impacts
4. Haven’t found an ROI for
consolidation
5. We balance consolidation and
islands depending on tiers of
service
Core/Edge is better for disk
1. The edge switch is not as highly available as a director-class switch, so why put the most expensive component, the disk frame, on an edge switch?
2. The CoLo SAN isolates disk frames within functional groupings of servers. This is akin to the limitation of direct-attached storage, except, in this case, a group of servers rather than a single server “owns” the storage.
3. The CoLo SAN presents other scalability issues such as the limitation of the number of ports in the edge switch.
5 Critical variables for sizing: SPICE
The “SPICE” variables for sizing Core/Edge SANs
S : How many SAN servers are needed?
P : How many regular servers will share a storage port?
I : How many regular servers will share an ISL between the edge switch and the director?
C : How many ports are on a director/core switch?
E : How many ports are on an edge switch?
P&I are most dependent upon your applications.
The SPICE variables
S How many SAN servers are needed?
S = 28
P = 7
I = 7
C = 140
E = 16
P How servers share a storage port?
I How servers share an ISL?
C How ports are on the director/core
switch?
E How ports are on the edge switch?
SPICE
The finer spicing
“P” = “I”
Why?
The goal is to fully utilize a storage port; therefore, the total bandwidth coming across the ISLs to that storage port will be equal to the bandwidth between the storage port and the director class switch. So, if 10 servers can fill a storage port pipe then they will also fill 1 ISL.
SPICE math for sizing*
S = Number of servers
P = I , or “PI”
Number of ISLs = ( S / PI )
Number of storage ports = ( S / PI )
Number of edge switches = (S + (S/PI) ) / E
Server capacity of core switch = ( C / 2 ) * PI
* Round up for each division
SPICEWhat is the practical effect of “PI”?
Helps with charge-back. Provides a metric for separating the hog servers from the regulars, and perhaps charging more for hogs.
It can be set higher than your Islands of SANs’ value but lower than what will probably be achieved. Thus, the migration can be properly budgeted and reports a moderately easy success. Following migration the production team can further refine the figure to a higher value.
SPICE for new Core/Edge SAN What is your S and P/I?
1. “S” is easy: How many servers do you want to have on the Core/Edge SAN when you declare a migration milestone? A question of project scope!
2. “PI” is harder • Use existing SAN Island as a baseline but you can probably
do better• Use storage utilization metrics from critical non-SAN
servers that will migrate• Rely on vendor’s experience• LOW estimates are easier to achieve
Comparative sizing
SPICE and 3 vendor comparisons
Brocade
Cisco
McData
SPICE for new Core/Edge SAN
PI = 7
What is your start point?Let’s assume:
Core/Edge SAN Building Block Brocade
Brocade 3900
E = 32 ports per edge switch
Brocade 24000
C = 128 ports per Core/Director switch
How many servers are supported by 1 Brocade edge switch?
Answer:
= ( (E / ( I + 1 ) ) * I
= ( (32 / ( 7 + 1 ) ) * 7
= 28 Servers
(see next slide)
How many servers are supported by 1 Brocade edge switch?
Answer = E - ( E / ( PI + 1 ) )
Answer = 32 – ( 32 / ( 7 + 1 ) )
Answer = 28
How many servers could 1 Brocade Director support with this SPICE?
Answer = ( C /2 ) * PI
Answer = ( 128 /2 ) * 7
Answer = 448
Cisco 9140
E = 40 ports per edge switch
Cisco 9509
C = 112 ports per Core/Director switch
Core/Edge SAN building block CISCO
A caveat
Fully-populated, the 9509 can hold 224 ports if 32-port blades are placed in all 7 slots. An assumption in my Core/Edge model is that you want to drive ISLs and storage points to maximum bandwidths which requires a non-blocking architecture.
The 32-port blades can be very useful for attaching lesser performing devices directly into the core, but in this case the core switch takes on roles that the Core/Edge model would delegate to the Edge switches.
Core/Edge SAN building block CISCO
C = 112 ports per Core/Director switch
How many servers are supported by 1 Cisco edge switch?
Answer:
= ( (E / ( I + 1 ) ) * I
= ( (40 / ( 7 + 1 ) ) * 7
= 35 Servers
(see next slide)
How many servers are supported by 1 Cisco edge switch?
Answer = E - ( E / ( PI + 1 ) )
Answer = 40 – ( 40 / ( 7 + 1 ) )
Answer = 35
How many servers could 1 Cisco Director support with this SPICE?
Answer = ( C /2 ) * PI
Answer = ( 112 /2 ) * 7
Answer = 392
McData 4500
E = 24 ports per edge switch
McData 6140
C = 140 ports per Core/Director switch
Core/Edge SAN building block McDATA
How many servers are supported by 1 McData edge switch?
Answer:
= ( (E / ( I + 1 ) ) * I
= ( (24 / ( 7 + 1 ) ) * 7
= 21 Servers
(see next slide)
How many servers are supported by 1 McData edge switch?
Answer = E - ( E / ( PI + 1 ) )
Answer = 24 – ( 24 / ( 7 + 1 ) )
Answer = 21
How many servers could 1 McData Director support with this SPICE?
Answer = ( C /2 ) * PI
Answer = ( 140 /2 ) * 7
Answer = 490
SAN math for a Core/Edge SANs
Conclusions• A Core/Edge SAN has advantages for disk SANs
• Sizing for a Core/Edge SAN is dependent on only 2 variables
under your control ( # of servers and PI or the fan-out ratio )
• Once you have determined your SAN goals and set these 2
variables, then you can work up a bill of materials from your
switch vendors rather than relying on their design/sales team