The (Almost) Definitive Guide to PowerFlex Sizing & Other Matters Flashcards

1
Q

What are the restrictions for software only PowerFlex?

A

only for existing ScaleIO customers (since most original ScaleIO deployments were SW only) or net new customers purchasing over 2 million in capacity licensing over 2 years

RPQ required

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

How much RAM is required per SDC?

A

50MB

also consumes very low host CPU

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

What is the SDSs responsibilities?

A

consuming local storage devices within server

making sure second copy of data is written to a different host before acknowledgement back to the SDCs

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

How much RAM does an SDS consume?

A

500MB-12GB per host (with MG layout)

very little CPU workload

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

How many vCPUs/cores are allocated for an SDS?

A

8-12

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What protection level do protection domains have?

A

only one simultaneous node failure can occur within a PD

fault sets allow for multiple host failures but are typically not required

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What should be the max number of SDSs to put in a PD?

A

30 SDSs per protection domain

to maintain 6 9s of availability

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

What is the max number of devices in a storage pool?

A

300 devices

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

How is data layed out in FG and MG pools?

A

MG - 1MB
FG - 4KB

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

What are the two performance profiles on PowerFlex?

A

High

Compact

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

What is the high performance profile?

A

all flash nodes

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

What is the compact performance profile?

A

HDD nodes

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

What is the SVM?

A

Storage VM

a VM for ESXi environments that runs CentOS - runs the SDS and LIA components

typically maps the local storage devices and RAID controller to the host through DirectPath IO

NVMe drives should use RDM

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

What is the main different between high and compact performance profile?

A

amount of CPU resources given

compact - SVM given 2 vCPUs/cores

high - SVM given 8 vCPUs/cores

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

What is the performance profile recommendation for PowerFlex?

A

configure High for both SSD and HDD - the new default setting

version 3.5 allows now up to 12 vCPUs and 12 SDS threads for FG pools

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

What is the recommendation specific to performance profile and CloudLink?

A

when using CloudLink D@RE increasing SVM by another vCPUs is considered best practice (from 8 to 12)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

What is the all-flash performance of PowerFlex?

A

each SSD can provide about 80,000 Read IOPS

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
18
Q

What is the max performance of SDS?

A

250,000 IOPS for reads
100,000 IOPS for writes (due to second copy being required it’s slower)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
19
Q

What happens after you hit the IOPS limit of an SDS?

A

can add more SSDs for higher large block performance

will not be able to improve small block performance

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
20
Q

What is the typical performance of an SDC?

A

500,000 read or write IOPS if connected to multiple SDSs

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
21
Q

What is the bandwidth performance of PowerFlex?

A

for large block IO - can saturate 200GbE per node to deliver 20GB/s of bandwidth for SDCs per node (2 x 100GbE interfaces)

SDS normally max out around 10GB/s per node when using NVMe and 7-8GB/s when using SSD

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
22
Q

What are the latency metrics of PowerFlex?

A

0.2-1ms w/ All Flash and very high IOPS

100-200 microseconds w/ NVMe and low queue depth IO profiles

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
23
Q

How does PowerFlex Manager run?

A

runs an an ova. VM on dedicated management nodes external to PowerFlex workload nodes

can also be managed through REST API

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
24
Q

What is a recommendation for sizing Oracle environments?

A

often beneficial to have compute-only nodes to extract the most value out of expensive licenses which are by the core

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
25
Q

What is the only node type that can support GPUs?

A

R740xd

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
26
Q

Why do you need to configure minimum 10% spare capacity?

A

to maintain protection of a single node failure

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
27
Q

What is the rule when sizing for performance?

A

the more nodes the better

size with smaller drives to get more nodes

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
28
Q

Why is bandwidth and latency better on NVMe devices rather than SSDs?

A

IO does not need to go through the HBA330 controller and instead goes directly to the CPU

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
29
Q

How is PowerFlex licensed?

A

per raw TiB - perpetual

can carry it over to net new hardware to reduce TCO

licenses per customer not per site or per cluster

30
Q

How many tiers of licensing are their?

A

11 Tiers

31
Q

What is tier 1 licensing?

A

12-100TBs

32
Q

What is tier 2 licensing?

A

101-250TBs

33
Q

What is tier 3 licensing?

A

251-500TBs

34
Q

What is tier 4 licensing?

A

501-1000TBs

35
Q

What is tier 5 licensing?

A

1001-5000TBs

36
Q

What is tier 6 licensing?

A

5001-10000TBs

37
Q

What is tier 7 licensing?

A

10001-20000TBs

38
Q

What is tier 8 licensing?

A

20001-40000TBs

39
Q

What is tier 9 licensing?

A

40001-60000TBs

40
Q

What is tier 10 licensing?

A

60001-100000TBs

41
Q

What is tier 11 licensing?

A

100000+TBs

42
Q

What is the RPO for asynchronous replication?

A

30 seconds for version 3.5

43
Q

How much RAM is required for SDR?

A

22GB RAM per node

44
Q

What is the connectivity recommendation for SDR?

A

25GbE or faster

45
Q

Where is replication enabled on PowerFlex?

A

at the protection domain level

SDRs must be present in each PD if any volumes from that PD need to be replicated

46
Q

What are the replication design choices for PowerFlex?

A

SDR only enabled on one PD

enable only Partial SDRs per PD

enable max nodes w/ SDR

47
Q

What is the SDR only enabled on one PD option?

A

allow only for volumes on that PD to be replicated

if this PD is down for maintenance replication will be unavailable during that time

48
Q

What is the enable only partial SDRs per PD?

A

will allow all PDs to be able to participate in replication activities

if the workload is low at least 3 nodes per PD need to be running the SDR

design may require an RPQ

49
Q

What is the enable max nodes with SDRs?

A

current supported max SDR per cluster is 64 (version 3.5)

50
Q

What is the relationship between WAN bandwidth and replication?

A

replication performance most often limited by the WAN bandwidth

need to properly size WAN bandwidth and journal capacities

51
Q

What is the relationship between local bandwidth and replication?

A

bandwidth consumption w/ replication is 3x higher than without which is why additional bandwidth is required

8K workload EX:
normal write = 16K (2 copies of 8K)
replicated write = 48K

52
Q

What is the networking recommendation for PowerFlex?

A

use multipathing whenever possible along with LACP

LACP is first layer of protection and then there are multiple VLANs running on top providing IP Multipathing

53
Q

What is important for OLTP workloads?

A

require high IOPS and low latency

54
Q

What is important for OLAP workloads?

A

high bandwidth

55
Q

How is SDS installed in a preconfigured ESXi environment?

A

deploying SVM

56
Q

How does the MDM interact with the SDS and SDC?

A

interacts asynchronously using a separate data path and will not impact performance

57
Q

What do larger storage pools provide?

A

wider striping by spreading IOs across more devices - benefits apps w/ high IO concurrency and bandwidth requirements

more flexible shared capacity for multiple volumes by providing larger pool for volume allocation

faster rebuild time - more spindles typically means faster rebuilds

58
Q

What do smaller storage pools provide?

A

better protection from multiple drive failures

better guaranteed IOPS by provisioning SPs for apps that require guaranteed performance

59
Q

How is physical and logical data protection done on PowerFlex?

A

PDs provide physical separation and SPs provide logical separation

SDCs can access volumes regardless of PD or SP

60
Q

What is important to know before implementing fault sets?

A

can only create and configure fault sets before adding SDSs to the system

configuring them incorrecty may prevent the creation of volumes since an SDS can only be added to a fault set during the creation of the SDS

61
Q

What are the rules for fault sets?

A

data mirroring for all the devices in the fault set must take place in a separate fault set

minimum of 3 fault sets are defined

62
Q

Where is spare capacity implemented in PowerFlex?

A

the SP level

63
Q

What is the relationship between rebuild/rebalance operations and bandwidth?

A

rebuild/rebalance can affect the available bandwidth for client traffic if the physical ports are shared between SDC and SDS nodes

64
Q

What is the networking recommendation when it comes to rebuild/rebalance operations?

A

Have multiple interfaces connected to allow front end client traffic to flow unimpeded by back end operations

back end rebuild/rebalance can compete w/ client bandwidth if only over a single interface

65
Q

How can interfaces be designated in PowerFlex?

A

SDC to SDS communication

SDS to SDS traffic

66
Q

What is the networking recommendation if you only have a single interface connection per node?

A

limit client bandwidth to keep from overloading system while a rebuild/rebalance occurs

67
Q

How can MDMs be mounted in PowerFlex?

A

can be located anywhere in the system and don’t require dedicated nodes

in larger configurations they should be distributed for availability reasons

EX: in multirack config they each (3-5) should be placed in different racks and different fault units

68
Q

What is ideal network configurations for MDM connectivity?

A

all on the same subnet for minimal latency

69
Q

What is the networking recommendation for smaller configurations?

A

running full mesh architecture w/ SDSs and SDCs sharing a non-routed subnet (access-agg)

70
Q

What is the networking recommendation for larger configurations?

A

leaf-spine

71
Q

Why should you size for multiple storage pools initially?

A

allows customer growth without having to provision multiple storage pools and keeps them balanced with respect to number of drives in them