Download presentation
Presentation is loading. Please wait.
Published byKevin Hutchinson Modified over 8 years ago
1
1 EMC CONFIDENTIAL—INTERNAL USE ONLY SMI-S Performance Overview and Status 8/28/2014
2
2 EMC CONFIDENTIAL—INTERNAL USE ONLY Agenda Objectives of Testing Overview of results Takeaways Action items
3
3 EMC CONFIDENTIAL—INTERNAL USE ONLY Objectives Answer the FAQ – what’s the supported scalability of SMIs dedicated for ViPR. Study performance characteristics of ViPR operation when the provider is managing a single VMAX Start with an empty VMAX and build it up to near max limit At every step capture the response time of common operations Study performance characteristics of ViPR operations on a pool of VMAX arrays managed by a single provider Combine single large and multiple (4) smaller VMAX arrays to arrive at max supported scalability of a provider
4
4 EMC CONFIDENTIAL—INTERNAL USE ONLY Environment (no Bound Resources) SMI-S Provider ESX Server 12 Core Intel Xeon CPU X5680 @3.33GHz 98 GB RAM Local disk Virtual Machine 2 vCPUs 16 GB Memory 50 GB Disk Space Inband connection to all VMAX though RDM with 6GK for each VMAX SMI-S Provider version: V4.6.2.11 Solutions Enabler version: V7.6-1808 2.0
5
5 EMC CONFIDENTIAL—INTERNAL USE ONLY Discovery and Metering Times at Various VMax Size (Single VMAX Single SMI-S Provider)
6
6 EMC CONFIDENTIAL—INTERNAL USE ONLY Metering and Monitoring CPU Character On SMI-S Metering and Discovery Do not appear to be CPU bound on SMI-S provider From logs ~ 90% of Time Metering and Discovery happen on SMI-S Provider Single 40 K VMax
7
7 EMC CONFIDENTIAL—INTERNAL USE ONLY Response time of common provisioning operations at various device count of a single VMAX array
8
8 EMC CONFIDENTIAL—INTERNAL USE ONLY Response time of common provisioning operations at various device count of a single VMAX array 300 Device Response time (sec) 9,300 Devices Response time (sec) 15,000 Devices Response time (sec) 25,000 Devices Response time (sec) 35,000 Devices Response time (sec) 40,000 Devices Response time (sec) CREATE_VOLUME 74 75 72 88 98 102 EXPAND_VOLUME 76 94 99 124 161 155 CREATE_VOLUME_SNAPSHOT 89 98 122 158 185 203 DEACTIVATE_VOLUME_SNAPSHOT 76 112 124 146 170 195 DEACTIVATE_VOLUME 59 118 142 160 185 203 CREATE_VOLUME_EXPORTGROUP 48 33 32 42 53
9
9 EMC CONFIDENTIAL—INTERNAL USE ONLY Time to create a single large devices as reported by ViPR Talk to Rohini- meta volume set to create large devices - Important
10
10 EMC CONFIDENTIAL—INTERNAL USE ONLY 40K Devices / 4 VMAX Metering Metering Monitoring happen mostly sequentially Avg Metering time – 12:41 (single large Vmax (28 minutes) Avg Discovery time – 8:31 (single large Vmax (25 minutes) Note only 4 GK were used on Vmax 195701185
11
11 EMC CONFIDENTIAL—INTERNAL USE ONLY ~40K Devices High Level View of CPU usage for SMI-S Discovery and Metering every 1 hour Single VMAX (~40K devices) being created using ViPR 4 VMAX arrays (~40K devices) - Ingested
12
12 EMC CONFIDENTIAL—INTERNAL USE ONLY Single Provider managing multiple VMAX arrays - observations Hourly Discovery Discovery runs in parallel per array Hourly Metering Metering runs sequentially per array
13
13 EMC CONFIDENTIAL—INTERNAL USE ONLY Throughput analysis of snap operations Total projected 1709 snaps per day for a single provider Total goal per provider with 5 Arrays 24K Assuming ~5 VMAX per provider and total 12 providers in full scale deployment 20,508 snaps per day are projected Required throughput for SNAP not in ball park
14
14 EMC CONFIDENTIAL—INTERNAL USE ONLY Response time analysis of snap operations Response time degrades modestly when a single thread goes to each VMAX. 4 VMAX 1 thread each show 13% degrade in response time Response time degrades significantly when multiple threads go to a single VMAX. 4 VMAX 3 threads each show a 2.8 fold degradation in response time
15
15 EMC CONFIDENTIAL—INTERNAL USE ONLY Multiple VMAX Snap Time on single provider 4Vmax Single Threaded 4Vmax 3 threads each
16
16 EMC CONFIDENTIAL—INTERNAL USE ONLY Multiple Vmax SMI-S Character 4Vmax Single Threaded4Vmax 3 threads each
17
17 EMC CONFIDENTIAL—INTERNAL USE ONLY Multiple VMax Snap Times as Reported by ViPR Important note – Above times are a different test run from previous slide and result can not be compared.
18
18 EMC CONFIDENTIAL—INTERNAL USE ONLY 80K Devices 5 VMax Environment Metering and Discovery Time Avg Metering time – 42:50 Avg Discovery time – 27:42
19
19 EMC CONFIDENTIAL—INTERNAL USE ONLY Hi level view of CPU utilization of SMI-S Provider for 80K+ Devices for VMax
20
20 EMC CONFIDENTIAL—INTERNAL USE ONLY Take Aways Lab environment - Meter/Discovery 83,000 Devices on a single SMI-S provider was a pass Support 60,000 Devices in field from Meter/Discover with out increasing polling interval Lab Environment Response time degrades significantly with on SMI-S provider when concurrent operations are sent to same VMax Lab Environment – Snap throughput not in the in the ball park regardless of concurrency or number of arrays per SMI-S provider Ameer comment - (Is a different method needed to create snap - open Jira for this)
22
22 EMC CONFIDENTIAL—INTERNAL USE ONLY Response time analysis of snap operations Provider managing multiple arrays Total projected 1709 snaps per day for a single provider Assuming ~4 VMAX per provider and total 7 providers in a deployment, 17,090 snaps per day are projected Assuming ~1 VMAX per provider and total 18 providers in a deployment, 17,090 snaps per day are projected 33 VMAX at 100% scale snaps per day goal is 128,300 7x improvement is required to meet goal
23
23 EMC CONFIDENTIAL—INTERNAL USE ONLY Agenda Meter Only 5 VMax total ~80 K
24
24 EMC CONFIDENTIAL—INTERNAL USE ONLY Agenda Meter Only
25
25 EMC CONFIDENTIAL—INTERNAL USE ONLY Agenda Meter Only
26
26 EMC CONFIDENTIAL—INTERNAL USE ONLY Single Poll 40K Volumes 4 Vmax Objectives of Testing Over View of results Take Away
27
27 EMC CONFIDENTIAL—INTERNAL USE ONLY 40K Devices 4 Vmax MeterMonitor 6 hours SMI-S Objectives of Testing Over View of results Take Away
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.