Forum:Would a hybrid hard drive (SHDD) improve performance for high IO bioinformatics under linux?

Question

[Daniel]

I'm running out of space on my work machine and intend on buying a new HDD drive to add to the 3TB HDD (storage) and 250GB SSD (OS and pseudo-scratch space) I already have.

I wonder whether the hybrid drives (i.e. this 4TB HDD and 8GB SSD) would have any benefit, as I don't really know how the drive would work with stuff like high throughput NGS IO. The standard way they work is by loading in regular applications into the SSD portion, but that's not really going to help with processing, and I already have my OS on the SSD.

Does anyone have any experience with these drives in their machines, and whether they actually help speed things up at all?

[Chris Miller]

Seems unlikely to have much of an impact on typical IO-heavy analyses to me. At the end of the day, the throughput is still limited by your ability to pull the data from that 100GB bam file, which will be on spinning disk, not the SSD.

That said, even for day-to-day work (email, browsing, etc), SSDs are amazing. Heavy apps, like Illustrator, no longer take minutes to open - they just pop open.

Answer 1

[Alastair Kerr]

If it is a storage drive like you said, then SSHD is unlikely to change the performance very much compared to a similar HDD for most applications. Exceptions might be serving the same file repeatedly to applications such as IGV using bigwig files or smaller BAM files. That said I've found that memory limits are the key issue when experiencing genome browser based problems and I have never had to investigate disk I/O issues.

For heavy disk I/O, two key factors are the speed of the drive (10000RPM - 7200RPM - 5400RPM) and whether the drives are on RAID. Even using a workstation, having your storage RAID 1 or RAID 6 will make a big difference for intensive disk operations and will provide better data security.

You may also want to check if disk I/O is even a problem. Is it reaching saturation when you are running your programs? I recommend looking at this guide.

[Daniel]

Thanks for this, great insight.