EFS has arrived Yaa! Yaa! Yaa! (I compared various Elastic File Systems)
table of contents
This is Sato from Beyond, whose title seems to reveal the generation.
Many years have passed since its release, and EFS (Elastic File System), which was announced at this year's AWS Summit, has finally been made compatible with the Tokyo region.
I was so happy that I summarized the things to consider before actually using the service.
How much does the cost change compared to EBS?
*Calculated at unit price per GB, 1$=110 yen | dollar | circle |
EFS | $0.36 | ¥40 |
EBS Provisioned IOPS SSD (io1) | $0.14 | ¥16 |
EBS general purpose SSD (gp2) | $0.12 | ¥13 |
EBS throughput optimized HDD (st1) | $0.05 | ¥6 |
EBS Cold HDD (sc1) | $0.03 | ¥3 |
Bonus S3 | $0.03 | ¥3 |
Since the number of types of EBS has increased, I decided to compare them all at once.
It's quite expensive... It's three times the cost compared to the most commonly used general-purpose SSD.
I'm a little hesitant when it comes to needing TB-level storage.
However, when you consider the effort and cost of creating redundant NFS instances for shared storage, you may be tempted to give in.
How good is EFS?
Verification environment
Instance: t2.micro
OS: AmazonLinux
EFS: Generic performance mode
I used the fio command to compare the performance of IOPS and throughput (MB/sec) with block sizes of 4K and 8M.
Reference: I measured storage performance using fio
First, let's compare IOPS.
I expected this to some extent since it's a system like NFS, but when you read and write a large number of small-sized files, the difference in IOPS becomes noticeable.
It was a little unexpected that random reads performed better than sequential reads, but this is an understandable result considering that EFS is designed as a decentralized distributed data storage
I think most of the actual use cases are random reads, so I'm glad that there is no big difference in read performance compared to local EBS.
Next, I compared the throughput (MB/sec)
The graph shows that there is not much difference in IOPS, but since it is NFS, the performance will be lower than EBS when writing small size (about 4K).
However, it seems that the overall throughput and IOPS of EFS improve as the average I/O size increases, so you may not need to worry too much.
This time I tested in general-purpose performance mode, but EFS has a maximum I/O performance mode, so if I have a chance, I would like to compare the performance differences between the different modes.