Sync limited speed

Software-based VM-centric and flash-friendly VM storage + free version

Moderators: anton (staff), art (staff), Max (staff), Anatoly (staff)

Post Reply
MohamadB
Posts: 4
Joined: Thu Jun 04, 2020 12:56 pm

Tue Nov 10, 2020 5:43 am

Dear All

I'm using vsan I have 2 nic cards for sync 10G and 2 nic for HB 10G when I start syncing between starwind1 and starwind2 I only get on each nic 1.9G/ps (SSD disk write speed 450MB/s)

when I mark 2 disks to sync speed become on every disk 250MB/s and the sync nic stay 1.9G/ps on both nic i used ipref3 and test was 10G/ps on every nic

so if I used 2 disks to sync and the speed split in half and the nic stay 1.9g/ps so starwind not using more then 1.9g/ps

any idea

vsphere enviroment back to back network intel network card jumbo frame 9000 enabled i used thick-provisioned star wind disk with l1 cach and l2 cach on ssd vsphere disk eagerd zerod all

Thanks,
yaroslav (staff)
Staff
Posts: 2279
Joined: Mon Nov 18, 2019 11:11 am

Tue Nov 10, 2020 6:24 am

Hi,

Welcome to StarWind Forum. May I learn the underlying storage configuration? What RAID level do you have? Here is how you should benchmark the underlying storage performance https://www.starwindsoftware.com/best-p ... practices/.
Also, are you reading in GB/s or Gbit/s?
May I learn what NICs do you have? I need the model.
Please consider disabling caches: I am not sure that you will get a significant performance boost with RAM and SSD cache on SSDs.
Are you using the same SSDs for L2 cache? If so, it will not make the performance any better. More info about caching for all-flash arrays https://knowledgebase.starwindsoftware. ... sh-arrays/.
Here is how to disable L1 and L2 caches respectively https://knowledgebase.starwindsoftware. ... dance/661/, https://knowledgebase.starwindsoftware. ... -l1-cache/.
MohamadB
Posts: 4
Joined: Thu Jun 04, 2020 12:56 pm

Tue Nov 10, 2020 6:21 pm

Hi

regarding L1 and L2 caches you are telling me that if i use SSD disk i dont need to use L1 and L2 right

My setup

group 1 : raid 1 480 SSD vSphere OS

* vSphere all switchs jumbo frame 9000

group 2 : raid 1 980 SSD

* Disk 1 = 150GB eager zeroed for StarWind
* Disk 2 = 150GB eager zeroed for StarWind

group 3 : raid 1.8TB SSD

* Disk 1 = 300GB eager zeroed for StarWind
* Disk 1 = 300GB eager zeroed for StarWind
* Disk 1 = 800GB eager zeroed for StarWind

=====================================================
Starwind1: All Thick provisioned With L1 And L2
8CPU
32GB RAM
* Disk 1 = 150GB L1 3GB L2 40GB WT 512
* Disk 1 = 150GB L1 3GB L2 40GB WT 512
* Disk 1 = 300GB L1 3GB L2 30GB WT 512
* Disk 1 = 300GB L1 3GB L2 30GB WT 512
* Disk 1 = 800GB L1 6GB L2 70GB WT 512

Network :

x2 sync 10G BACK TO BACK jumbo frame 9000 (vSphere)
x1 HB 10G BACK TO BACK jumbo frame 9000 (vSphere)
x1 HB 10G BACK TO BACK jumbo frame 9000 (direct to operating system using iSCSI Init)
x1 HB 10G BACK TO BACK jumbo frame 9000 (direct to operating system using iSCSI Init)

Server 2 have the same configuration

Network Adapter X550-T2
using ipref3 from starwind1 to starwind2 for all the 5 10G NICS

I have {SUM} 11.9 Gbytes 10.1Gbits/sec

CrystalDiskMark in starwind1 and starwind2

I have read 2000MB/S and write 450MB/S

i stored the L2 cache in raid 1 which i used for the OS
yaroslav (staff)
Staff
Posts: 2279
Joined: Mon Nov 18, 2019 11:11 am

Tue Nov 10, 2020 7:37 pm

Greetings,
regarding L1 and L2 caches you are telling me that if i use SSD disk i dont need to use L1 and L2 right
I am just saying that it might not bring significant performance gain.

Are you using the software RAID?
How large is the test file?
May I wonder how do you measure synchronization speed? What you can do is measuring performance with 64K sequential and random writes/reads - the pattern that simulates the synchronization as discussed here https://www.starwindsoftware.com/best-p ... practices/
MohamadB
Posts: 4
Joined: Thu Jun 04, 2020 12:56 pm

Sat Nov 14, 2020 9:01 am

hi

I'm using hardware raid RAID Controller RS3DC080
Test file size 1GB using Crystal Disk Mark (disk full write speed was 450MB)
how did i know that synchronization speed only 1.9Gbps i did not used any tool i used task manager performance tab
during the sync i noticed that my sync link is 1.9Gbps max i can't see 2Gbps only 1.9Gbps and i have 2 sync links every link is 1.9Gbps

and i was syncing only (1) HA device image size 300GB using task manager performance tab i check the disk tab for HA image 300Gb and i found the disk write speed is 465MB/s
i selected anther HA device image 300GB from starwind and i clicked sync this HA image located on different disk (the write speed changed from 465MB/s to 235MB/s on the first ha disk which have 300GB image)
(and on the second disk write speed also became something like 230MB/s ) and the sync speed on both link did not change from 1.9Gbps ----

i think logic say (speed on disk must stay 450MB/s on every disk must not change ) and the sync speed should become 3.9Gbps on every sync link

all disk are SSD

now im building anther cluster and i will not use L1 or L2 cache and check the speed
yaroslav (staff)
Staff
Posts: 2279
Joined: Mon Nov 18, 2019 11:11 am

Sat Nov 14, 2020 4:46 pm

Could you tell me what is the RAID level you use? Did you benchmark the underlying storage according to the article I shared before?

Please note that 2 Sync links are working as round robin but not as a team. In other words, 2 Sync links are more about network redundancy but not performance boost.

Please see if sync speed changes without cache.
jamecorbitt
Posts: 1
Joined: Thu Jul 21, 2022 1:05 pm
Contact:

Thu Jul 21, 2022 1:15 pm

this fix would be very helpful. I will be glad if there is news about it :)
yaroslav (staff)
Staff
Posts: 2279
Joined: Mon Nov 18, 2019 11:11 am

Thu Jul 21, 2022 2:14 pm

Stay tuned! The release notes are available at https://www.starwindsoftware.com/release-notes-build
Post Reply