EMC ScaleIO – these guys won’t miss a chance to miss a chance: testing EMC ScaleIO 4-Node Cluster with Performance profile set to High performance

Cult of Anarchy headquarters. 7 months earlier.

  • What are we testing today?
  • EMC ScaleIO, Sir!
  • Fcop, this rocket must fly! How many VMs per a single node this beast gives us?
  • ONE…
  • What the hwem?! How many IOPS???
  • …we tried to add more VMs, Sir, but things just got worth…the max was 94000 IOPS…so we cloned a single VM to 4 cluster nodes…
  • Give me numbers!
  • 22% performance improvement…
  • Lord help us…

After the results of the first benchmark for EMC ScaleIO were published on our blog, the readers started arguing “What about enabling DM, SDC, and SDS for EMC ScaleIO you fkemheads?” Fair enough, enabling a High Performance profile on EMC ScaleIO might really boost its performance to the extent were Microsoft and StarWind (current performance leaders) become just a bunch of eqemgcvgtu constantly playing with themselves and their miserable IOPS. Well, we sure wanna give EMC ScaleIO one more chance to miss a chance 😊

All engines on! Re-investigating EMC ScaleIO performance

We have repeated our series of tests, this time, MDM, SDS, SDC were configured for ULTIMATE performance:

wp-image-1856

OK, let’s make sure we’re on the same page when it comes to terms.

Metadata Manager (MDM). It’s main goal is to configure and monitor the EMC ScaleIO system. It contains all the metadata required for system operation. Useless ujkv? Maybe, but it is responsible for data migration, rebuilds, and all system-related functions.

ScaleIO Data Server (SDS). This option manages the capacity of a single host and acts as a back-end for data access. The SDS is installed on all servers provisioning storage devices to the EMC ScaleIO system.

ScaleIO Data Client (SDC). A driver that provisions EMC ScaleIO volumes as block devices to all your apps located on the same server with SDC.

So checkboxes are there, but what about some real proof?

Let’s log in to the Master MDM (VM) via SSG and execute the following command:

scli –query_performance_parameters

wp-image-1857

OK guys, we did our best. The performance profile is set to high performance.

We decided not to bore our readers with all those network interconnection diagrams, system specs, and testing methodology since we already had this all described in our previous article right here: DELL EMC SCALEIO – AN “ENTERPRISE-GRADE” SOLUTION WITH AWFUL SCALABILITY

Picking optimal VM configuration

Since the methodolgy remains the same, we will first determine which vCPU/vCORE combination makes our VM fly.

Testing EMC ScaleIO 4-Node Cluster with 1xVM (2CPU/1CorePerSocket):

2 CPU/1 Cores per Socket Virtual Disk 1 – 620GB
DiskSPD (high performance setting) 4k random read (2 thread) Fio
(high performance setting)
4k random read (2 thread)
Test time 360sec Test time 360sec
  IOPS MB/s Latency (ms)   IOPS MB/s Latency (ms)
QD=1 7322 29 0,28 QD=1 6426 25 0,30
QD=2 13444 53 0,30 QD=2 12435 49 0,31
QD=4 24896 97 0,34 QD=4 22795 89 0,34
QD=6 36114 141 0,36 QD=6 32138 126 0,36
QD=8 45003 176 0,39 QD=8 39624 155 0,39
QD=10 51022 199 0,41 QD=10 47092 184 0,41
QD=12 57528 225 0,43 QD=12 55013 215 0,42
QD=14 60348 236 0,45 QD=14 60990 238 0,44
QD=16 72963 285 0,46 QD=16 68133 266 0,45
QD=18 76489 299 0,48 QD=18 74114 290 0,47
QD=20 80214 313 0,49 QD=20 79439 310 0,48
QD=24 101056 395 0,52 QD=24 91035 356 0,50
QD=28 112621 440 0,55 QD=28 100993 395 0,53
QD=30 132435 517 0,56 QD=30 135015 527 0,54
QD=32 165056 645 0,58 QD=32 162431 635 0,56
QD=64 201015 738 1,11 QD=64 185648 725 1,04
QD=128 212546 830 2,23 QD=128 208701 815 2,06

wp-image-1858

Testing EMC ScaleIO 4-Node Cluster with 1xVM (4CPU/1CorePerSocket):

4 CPU/1 Cores per Socket Virtual Disk 1 – 620GB
DiskSPD (high performance setting) 4k random read (4 thread) Fio
(high performance setting)
4k random read (4 thread)
Test time 360sec Test time 360sec
  IOPS MB/s Latency (ms)   IOPS MB/s Latency (ms)
QD=1 14452 56 0,30 QD=1 13885 54 0,31
QD=2 24627 96 0,34 QD=2 23805 93 0,34
QD=4 42827 167 0,39 QD=4 47581 186 0,39
QD=6 75016 293 0,44 QD=6 74566 291 0,43
QD=8 117845 460 0,47 QD=8 118874 464 0,46
QD=10 148930 582 0,51 QD=10 145536 569 0,50
QD=12 175568 686 0,54 QD=12 172889 675 0,52
QD=14 206233 806 0,57 QD=14 204426 799 0,55
QD=16 228014 891 0,61 QD=16 217683 850 0,58
QD=32 215473 842 1,18 QD=32 212005 828 1,17
QD=64 216268 845 2,39 QD=64 212148 829 2,26
QD=128 215615 842 4,74 QD=128 216322 845 4,56

wp-image-1859

Testing EMC ScaleIO 4-Node Cluster with 1xVM (6CPU/1CorePerSocket):

6 CPU/1 Cores per Socket Virtual Disk 1 – 620GB
DiskSPD (high performance setting) 4k random read (6 thread) Fio
(high performance setting)
4k random read (6 thread)
Test time 360sec Test time 360sec
  IOPS MB/s Latency (ms)   IOPS MB/s Latency (ms)
QD=1 18326 72 0,33 QD=1 17607 69 0,33
QD=2 32881 128 0,36 QD=2 32065 125 0,36
QD=4 56302 220 0,43 QD=4 53485 209 0,44
QD=6 101223 395 0,49 QD=6 72608 284 0,48
QD=8 134478 525 0,53 QD=8 128963 504 0,54
QD=10 172525 674 0,58 QD=10 170166 665 0,57
QD=12 212043 828 0,64 QD=12 206855 808 0,64
QD=14 213424 834 0,73 QD=14 209003 816 0,72
QD=16 211319 825 0,84 QD=16 211724 827 0,84
QD=32 210043 820 1,71 QD=32 210410 822 1,72
QD=64 209566 819 3,35 QD=64 210402 822 3,37
QD=128 208336 814 6,70 QD=128 207008 809 6,75

wp-image-1860

As we all can see, the maximum performance of the tested disk in a single VM is almost independent of the CPU number and equals 216000-217000 IOPS. IMPRESSIVE! Correspondingly, taking the maximum value we managed to squeeze, we’re gonna use the 4 CPU/1CorePerSocket VM configuration in all our further tests.

More VMs, more performance – testing several VMs combined performance on a single node

OK, now that we have a perfect configuration for our VMs, let’s increase their number, keeping them pinned to a single node. We will stop only after their overall performance refuses to grow. Here are the configurations we used in our benchmarks for EMC ScaleIO 4-Node Cluster:

1. 2xVM (4CPU/1CorePerSocket/Virtual Disk1=310GB)

2. 3xVM (4CPU/1CorePerSocket/Virtual Disk1=210GB)

Remember, we’ve got High performance profile set ON so we won’t see the performance drop any time soon, right? WRONG!

Testing EMC ScaleIO 4-Node Cluster 2xVM (4CPU/1CorePerSocket/Virtual Disk1=310GB):

2xVM (4CPU/1 Cores per Socket) Virtual Disk 1 – 310GB
DiskSPD (high performance setting) 4k random read (4 thread) Fio

(high performance setting)
4k random read (4 thread)
Test time 360seс Test time 360sec
VM1 IOPS MB/s Latency (ms) VM1 IOPS MB/s Latency (ms)
QD=1 11347 44 0,35 QD=1 8738 34 0,45
QD=2 19505 76 0,41 QD=2 13270 52 0,59
QD=4 34333 134 0,47 QD=4 21548 84 0,73
QD=6 47886 187 0,50 QD=6 30210 118 0,78
QD=8 60038 235 0,57 QD=8 40185 157 0,78
QD=10 71564 280 0,65 QD=10 55349 216 0,71
QD=12 80049 313 0,71 QD=12 67278 263 0,70
QD=14 88601 346 0,77 QD=14 80467 314 0,68
QD=16 94696 370 0,84 QD=16 90464 353 0,69
QD=32 118048 461 1,16 QD=32 121493 475 1,04
QD=64 117980 461 2,25 QD=64 115305 450 2,20
QD=128 117425 459 4,41 QD=128 120475 471 4,23
VM2 IOPS MB/s Latency (ms) VM2 IOPS MB/s Latency (ms)
QD=1 12151 47 0,35 QD=1 8478 33 0,46
QD=2 19189 75 0,41 QD=2 13059 51 0,60
QD=4 34163 133 0,47 QD=4 21324 83 0,74
QD=6 46539 182 0,50 QD=6 30171 118 0,78
QD=8 61102 239 0,57 QD=8 41036 160 0,77
QD=10 71277 278 0,65 QD=10 54215 212 0,73
QD=12 79677 311 0,72 QD=12 67275 263 0,70
QD=14 87462 342 0,79 QD=14 80468 314 0,68
QD=16 93249 364 0,86 QD=16 91524 358 0,68
QD=32 119057 465 1,18 QD=32 116998 457 1,08
QD=64 117200 458 2,26 QD=64 116288 454 2,18
QD=128 118001 461 4,43 QD=128 121884 476 3,98

wp-image-1861

Testing EMC ScaleIO 4-Node Cluster 3xVM (4CPU/1CorePerSocket/Virtual Disk1=210GB):

3xVM (4CPU/1 Cores per Socket) Virtual Disk 1 – 210GB
DiskSPD (high performance setting) 4k random read (4 thread) Fio
(high performance setting)
4k random read (4 thread)
Test time 360sec Test time 360sec
VM1 IOPS MB/s Latency (ms) VM1 IOPS MB/s Latency (ms)
QD=1 10282 40 0,39 QD=1 10195 40 0,38
QD=2 17680 69 0,45 QD=2 16332 64 0,48
QD=4 31361 123 0,51 QD=4 26465 103 0,59
QD=6 43767 171 0,55 QD=6 34890 136 0,68
QD=8 53592 209 0,60 QD=8 44548 174 0,70
QD=10 60754 237 0,66 QD=10 57809 226 0,68
QD=12 65835 257 0,73 QD=12 63772 249 0,74
QD=14 70203 274 0,80 QD=14 69132 270 0,79
QD=16 69782 273 0,92 QD=16 72311 282 0,87
QD=32 75098 293 1,70 QD=32 80546 315 1,48
QD=64 75528 295 3,39 QD=64 73285 286 3,47
QD=128 73680 288 6,62 QD=128 71793 280 6,61
VM2 IOPS MB/s Latency (ms) VM2 IOPS MB/s Latency (ms)
QD=1 10402 41 0,38 QD=1 9984 39 0,39
QD=2 17761 69 0,45 QD=2 16096 63 0,49
QD=4 31427 123 0,51 QD=4 26190 102 0,60
QD=6 43846 171 0,55 QD=6 34752 136 0,68
QD=8 53744 210 0,59 QD=8 45010 176 0,70
QD=10 60990 238 0,66 QD=10 56332 220 0,70
QD=12 66191 259 0,72 QD=12 63869 250 0,74
QD=14 70602 276 0,79 QD=14 69072 270 0,79
QD=16 70235 274 0,91 QD=16 73304 286 0,86
QD=32 75755 296 1,69 QD=32 73698 288 1,62
QD=64 76280 298 3,36 QD=64 73874 289 3,45
QD=128 74480 291 6,55 QD=128 73920 289 6,43
VM3 IOPS MB/s Latency (ms) VM3 IOPS MB/s Latency (ms)
QD=1 10384 41 0,38 QD=1 9859 39 0,40
QD=2 17789 69 0,45 QD=2 15929 62 0,49
QD=4 31519 123 0,51 QD=4 25933 101 0,61
QD=6 44017 172 0,54 QD=6 34649 135 0,68
QD=8 53988 211 0,59 QD=8 45551 178 0,69
QD=10 61273 239 0,65 QD=10 55271 216 0,71
QD=12 66675 260 0,72 QD=12 63749 249 0,74
QD=14 71207 278 0,79 QD=14 69073 270 0,79
QD=16 70982 277 0,90 QD=16 75691 296 0,83
QD=32 77583 303 1,65 QD=32 70758 276 1,68
QD=64 79021 309 3,24 QD=64 74397 291 3,42
QD=128 76841 300 6,36 QD=128 77748 304 5,70

wp-image-1862

The increase of the number of VMs on a single node doesn’t give us a significant overall performance increase for the ScaleIO Volume. To explain this ridiculous situation, we provide you with a phone call script from one of our secret agents and EMC ScaleIO support:

– Hello, you have contacted EMC ScaleIO support, how may we help you?

– Hi! Well, I have set one VM on an EMC ScaleIO node and it runs really good, but once I increased the VMs number, the overall Volume performance didn’t really increase. Could you please help me with this?

– … (silence)

– Hello? Hello!? HELLO?!!?

Actually, the overall ScaleIO Volume performance with 3 VMs was lower than testing it with 2 VMs. Unfortunately, the increase of VMs number on a single node has no further point.

Now, we will clone the 2xVM (4CPU/1CorePerSocket) configuration to the remaining 3 cluster nodes (2 VMs per node).

Let’s dot the i’s and cross the t’s – testing if EMC ScaleIO actually scales

What we gonna do next, is test the overall 4-node EMC ScaleIO cluster performance with two VMs running on a single host simultaneously (8xVM/4CPU/1CorePerSocket /Virtual Disk1=80GB).We have reduced the Virtual Disk size to 80GB since the overall ScaleIO Datastore volume is constant and has to be split across 8 VMs.

Testing EMC ScaleIO 4-Node Cluster (8xVM/4CPU/1CorePerSocket /Virtual Disk1=80GB):

8xVM (4CPU/1 Cores per Socket)
Node1 Virtual Disk 1 – 80GB Node2 Virtual Disk 1 – 80GB
DiskSPD (high performance setting) 4k random read (4 thread) Fio
(high performance setting)
4k random read (4 thread) DiskSPD
(high performance setting)
4k random read (4 thread) Fio
(high performance setting)
4k random read (4 thread)
Test time 360sec Test time 360sec Test time 360sec Test time 360sec
VM1 IOPS MB/s Latency (ms) VM1 IOPS MB/s Latency (ms) VM1 IOPS MB/s Latency (ms) VM1 IOPS MB/s Latency (ms)
QD=1 9021 35 0,44 QD=1 8690 34 0,45 QD=1 9183 36 0,43 QD=1 8595 34 0,45
QD=2 15441 60 0,52 QD=2 14599 57 0,53 QD=2 15207 59 0,52 QD=2 14695 57 0,53
QD=4 22795 89 0,70 QD=4 22985 90 0,68 QD=4 22362 87 0,71 QD=4 22476 88 0,70
QD=6 24762 97 0,97 QD=6 24422 95 0,97 QD=6 24391 95 0,98 QD=6 25011 98 0,95
QD=8 25516 100 1,25 QD=8 25111 98 1,26 QD=8 25179 98 1,27 QD=8 24707 97 1,28
QD=10 25661 100 1,56 QD=10 25413 99 1,56 QD=10 25376 99 1,58 QD=10 26668 104 1,49
QD=12 25673 100 1,87 QD=12 25323 99 1,88 QD=12 25819 101 1,86 QD=12 25801 101 1,85
QD=14 26114 102 2,14 QD=14 25945 101 2,14 QD=14 26528 104 2,11 QD=14 26176 102 2,12
QD=16 26695 104 2,40 QD=16 26395 103 2,40 QD=16 27054 106 2,37 QD=16 26486 103 2,40
QD=18 26670 104 2,70 QD=18 34782 136 2,08 QD=18 26957 105 2,67 QD=18 28043 110 2,56
QD=20 27049 106 2,96 QD=20 26669 104 2,98 QD=20 27352 107 2,92 QD=20 27085 106 2,94
QD=24 27216 106 3,53 QD=24 27088 106 3,53 QD=24 27325 107 3,51 QD=24 27004 105 3,54
QD=28 27150 106 4,12 QD=28 27526 108 4,05 QD=28 27261 106 4,11 QD=28 27622 108 4,04
QD=30 27398 107 4,38 QD=30 28096 110 4,25 QD=30 27477 107 4,37 QD=30 28020 109 4,26
QD=32 27499 107 4,65 QD=32 27470 107 4,64 QD=32 27536 108 4,65 QD=32 27413 107 4,65
QD=64 27758 108 9,22 QD=64 27647 108 9,24 QD=64 27858 109 9,19 QD=64 27570 108 9,26
QD=128 27644 108 18,52 QD=128 27109 106 18,87 QD=128 27845 109 18,39 QD=128 27144 106 18,84
VM2 IOPS MB/s Latency (ms) VM2 IOPS MB/s Latency (ms) VM2 IOPS MB/s Latency (ms) VM2 IOPS MB/s Latency (ms)
QD=1 9032 35 0,44 QD=1 8483 33 0,46 QD=1 9203 36 0,43 QD=1 8445 33 0,46
QD=2 15477 60 0,52 QD=2 14594 57 0,53 QD=2 15225 59 0,52 QD=2 14768 58 0,53
QD=4 22818 89 0,70 QD=4 22752 89 0,69 QD=4 22345 87 0,71 QD=4 22054 86 0,71
QD=6 24799 97 0,97 QD=6 24199 95 0,98 QD=6 24365 95 0,98 QD=6 24443 95 0,97
QD=8 25531 100 1,25 QD=8 24982 98 1,27 QD=8 25161 98 1,27 QD=8 24746 97 1,28
QD=10 25629 100 1,56 QD=10 25509 100 1,55 QD=10 25383 99 1,58 QD=10 26262 103 1,51
QD=12 25654 100 1,87 QD=12 25503 100 1,86 QD=12 25790 101 1,86 QD=12 25781 101 1,85
QD=14 26138 102 2,14 QD=14 26019 102 2,13 QD=14 26540 104 2,11 QD=14 25858 101 2,15
QD=16 26739 104 2,39 QD=16 26453 103 2,41 QD=16 27042 106 2,37 QD=16 26547 104 2,39
QD=18 26647 104 2,70 QD=18 26489 103 2,70 QD=18 26943 105 2,67 QD=18 29314 115 2,49
QD=20 27083 106 2,95 QD=20 26712 104 2,98 QD=20 27383 107 2,92 QD=20 26885 105 2,96
QD=24 27223 106 3,53 QD=24 27004 105 3,54 QD=24 27318 107 3,51 QD=24 27071 106 3,52
QD=28 27172 106 4,12 QD=28 27515 107 4,05 QD=28 27265 107 4,11 QD=28 27668 108 4,03
QD=30 27426 107 4,38 QD=30 28386 111 4,21 QD=30 27505 107 4,36 QD=30 28073 110 4,25
QD=32 27500 107 4,65 QD=32 28590 112 4,47 QD=32 27574 108 4,64 QD=32 27405 107 4,65
QD=64 27756 108 9,22 QD=64 27661 108 9,23 QD=64 27883 109 9,18 QD=64 27574 108 9,26
QD=128 27645 108 18,52 QD=128 27345 107 18,71 QD=128 27957 109 18,31 QD=128 27097 106 18,88
Node3 Virtual Disk 1 – 80GB Node4 Virtual Disk 1 – 80GB
DiskSPD (high performance setting) 4k random read (4 thread) Fio
(high performance setting)
4k random read (4 thread) DiskSPD (high performance setting) 4k random read (4 thread) Fio
(high performance setting)
4k random read (4 thread)
Test time 360sec Test time 360sec Test time 360sec Test time 360sec
VM1 IOPS MB/s Latency (ms) VM1 IOPS MB/s Latency (ms) VM1 IOPS MB/s Latency (ms) VM1 IOPS MB/s Latency (ms)
QD=1 8822 34 0,45 QD=1 8546 33 0,46 QD=1 8868 35 0,45 QD=1 8524 33 0,46
QD=2 14979 59 0,53 QD=2 14996 59 0,52 QD=2 15161 59 0,53 QD=2 14590 57 0,53
QD=4 21943 86 0,73 QD=4 22221 87 0,71 QD=4 22212 87 0,72 QD=4 22613 88 0,69
QD=6 23957 94 1,00 QD=6 24971 98 0,95 QD=6 24251 95 0,99 QD=6 24356 95 0,97
QD=8 24823 97 1,29 QD=8 25137 98 1,26 QD=8 25050 98 1,28 QD=8 25061 98 1,26
QD=10 25054 98 1,60 QD=10 24983 98 1,58 QD=10 25255 99 1,58 QD=10 25132 98 1,58
QD=12 25162 98 1,91 QD=12 25827 101 1,85 QD=12 25354 99 1,89 QD=12 25298 99 1,88
QD=14 25691 100 2,18 QD=14 26175 102 2,12 QD=14 25763 101 2,17 QD=14 26376 103 2,11
QD=16 26283 103 2,43 QD=16 26909 105 2,36 QD=16 26382 103 2,43 QD=16 26481 103 2,40
QD=18 26457 103 2,72 QD=18 26495 104 2,70 QD=18 26458 103 2,72 QD=18 27261 106 2,63
QD=20 26929 105 2,97 QD=20 26765 105 2,97 QD=20 26928 105 2,97 QD=20 26757 105 2,97
QD=24 27194 106 3,53 QD=24 27133 106 3,52 QD=24 27110 106 3,54 QD=24 27066 106 3,53
QD=28 27227 106 4,11 QD=28 27713 108 4,03 QD=28 27212 106 4,12 QD=28 27522 108 4,05
QD=30 27469 107 4,37 QD=30 28019 109 4,26 QD=30 27355 107 4,39 QD=30 28051 110 4,26
QD=32 27591 108 4,64 QD=32 27484 107 4,63 QD=32 27539 108 4,65 QD=32 28007 109 4,56
QD=64 28290 111 18,10 QD=64 27599 108 9,26 QD=64 27845 109 9,19 QD=64 27567 108 9,26
QD=128 27992 109 9,14 QD=128 27229 106 18,78 QD=128 27570 108 18,57 QD=128 27352 107 18,70
VM2 IOPS MB/s Latency (ms) VM2 IOPS MB/s Latency (ms) VM2 IOPS MB/s Latency (ms) VM2 IOPS MB/s Latency (ms)
QD=1 8869 35 0,45 QD=1 8514 33 0,46 QD=1 8843 35 0,45 QD=1 8586 34 0,45
QD=2 15107 59 0,53 QD=2 14922 58 0,52 QD=2 15115 59 0,53 QD=2 14556 57 0,53
QD=4 22154 87 0,72 QD=4 22705 89 0,69 QD=4 22149 87 0,72 QD=4 22803 89 0,69
QD=6 24193 95 0,99 QD=6 24786 97 0,96 QD=6 24193 95 0,99 QD=6 24252 95 0,97
QD=8 25012 98 1,28 QD=8 25524 100 1,24 QD=8 25005 98 1,28 QD=8 24855 97 1,27
QD=10 25241 99 1,58 QD=10 25511 100 1,55 QD=10 25194 98 1,59 QD=10 25137 98 1,57
QD=12 25320 99 1,89 QD=12 25823 101 1,85 QD=12 25274 99 1,90 QD=12 25467 99 1,87
QD=14 25822 101 2,17 QD=14 26376 103 2,11 QD=14 25730 101 2,18 QD=14 26183 102 2,12
QD=16 26406 103 2,42 QD=16 26905 105 2,36 QD=16 26331 103 2,43 QD=16 26474 103 2,40
QD=18 26590 104 2,71 QD=18 26572 104 2,69 QD=18 26446 103 2,72 QD=18 26586 104 2,69
QD=20 27030 106 2,96 QD=20 26863 105 2,96 QD=20 26897 105 2,97 QD=20 26745 104 2,97
QD=24 27314 107 3,51 QD=24 27267 107 3,50 QD=24 27095 106 3,54 QD=24 27317 107 3,50
QD=28 27313 107 4,10 QD=28 27636 108 4,04 QD=28 27201 106 4,12 QD=28 27597 108 4,04
QD=30 27559 108 4,35 QD=30 28023 109 4,27 QD=30 27377 107 4,38 QD=30 28223 110 4,23
QD=32 27673 108 4,63 QD=32 29135 114 4,38 QD=32 27593 108 4,64 QD=32 28570 112 4,47
QD=64 28028 109 9,13 QD=64 27587 108 9,26 QD=64 27892 109 9,18 QD=64 27584 108 9,26
QD=128 28496 111 17,97 QD=128 27266 107 18,76 QD=128 27684 108 18,50 QD=128 27321 107 18,72

wp-image-1863

wp-image-1864

RIP

So we tried to resurrect the THING that was burried long ago deep beneath the ground…EMC ScaleIO. Result? It managed to stick out its one hand above the ground just to show everyone a middle finger. With high performance settings enabled in Performance profile of the EMC ScaleIO system, the test disk performance gets higher indeed…with…ONE…VM…running. Without high performance settings, we have managed to get 94000 IOPS, with high performance settings on – 222000 IOPS. Also, the increase of CPU number from 4 to 6 for our testing VM did not give us any speed boost.

But the main thing is that there is almost no performance increase while scaling EMC ScaleIO. After cloning 2 VMs (4CPU/1CorePerSocket) to the rest of cluster nodes, the benchmark showed that the overall cluster performance equals 2 VMs performance on a single ESXi host. If we compare EMC ScaleIO speed to that of a single Samsung SSD 960 EVO M.2 NVMe drive, EMC ScaleIO goes slower by 30000-32000 IOPS.

There are no NVMe drives enough to bring EMC ScaleIO back to life. Officially, EMC ScaleIO should be renamed to “EMC FailIO”. \

4.38/5 (13)

Please rate this