Posts

Hive-Engine Witness Node IOPS Upgrade

avatar of @ervin-lemark
25
@ervin-lemark
·
0 views
·
2 min read

As some of you know I am trying to do my part and help the platform as a Hive-Engine Witness.

Since the goal is also to have witness nodes geographically distributed as much as possible I choose a provider in Latvia. OK, it is cheap too :)

So, Time4VPS (referral link) offers VPS servers only. With, for example, 4 CPU 8 GB RAM server for 150 € for two years lease. Cheap, I told you.

Where is the catch? IO and disk speed are! Throttling is in place. For this price IOPS (input/output operations per second) is limited to 200. Which is by far not enough when you need to restore the database. 11 GB of archived database dump takes 25 hours to restore.

The solution? Change provider or ...? Luckily, and probably because I am not the first one hitting the limit, there is an upgrade option of increasing IOPS 5 times. Yes, a whopping 1,000 IOPS. Will this be enough? The price is 12 € per year. Cheap again.

This morning I bought the upgrade, the server restarted and after MongoDB cleanup, I started mongorestore. Here are some figures to show what went on.


Database utilization

The second peak of the Disk Utilization graph shows the speed test below.


IOPS today and this week

Quite telling by itself, isn't it? During the restore, teh average IOPS was around 760.


Disk speed tests

time dd if=hsc_08-16-2021_b56595416.archive of=/dev/null 
23187152+1 records in 
23187152+1 records out 
11871822060 bytes (12 GB, 11 GiB) copied, 105.641 s, 112 MB/s 
 
real    1m45.644s 
user    0m13.201s 
sys     0m35.109s 
time dd if=hsc_08-16-2021_b56595416.archive of=/dev/null bs=4M 
2830+1 records in 
2830+1 records out 
11871822060 bytes (12 GB, 11 GiB) copied, 47.5944 s, 249 MB/s 
 
real    0m47.598s 
user    0m0.013s 
sys     0m13.030s 

Quoting @forkyishere:

The first one tests very low block sizes, usually normal on most filesystems. The later one tests bandwidth mostly, a read test. What you need anyway since write is mostly buffered.

According to data from other witnesses, my node is twice as slow when it comes to bulk disk operations.

Will it be good enough to keep the pace and sign the blocks when needed? We'll see that in the near future.

For now, I will stick with Time4VPS (referral link) and hope that the server will serve its purpose. After all, when you don't have high expectations, their offering is very attractive.

Later, I will migrate to more powerful and more expensive hosting. How much later? We will see ...

Ps: if you wish to vote for me as a Hive-Engine witness use this link, please - https://votify.vercel.app/ervin-lemark. You will need some $WORKERBEE tokens for your vote to have influence.

Excellent Hive-Engine witness tools by @primersion.

Official witness page

@forkyishere and @reazuliqbal, thank you very much for your help.


Better and better