0

We have just ordered a new webserver with 120 Gb solid state disk and a SATA disk. I am trying to plan ahead what sort of filesystem to use.

This system will be running Linux, Apache/Tomcat to host java services. The main service is a system where people can upload reasonably large files (in the order of 100 Mb, images, image stacks and video), which people will be able to annotate and which will be sent to a database server when annotation is complete.

Thus far, I plan to put most of the utility programs of the operating system om the SSD and put the large media files there. The SATA disks will hold the less volitile data like apache, tomcat and the servlets.

For filesystems I have considered going for the stable EXT3 because I hear that it is best supported. The downside seems to be that it not the ideal choice for large files. That is why I am leaning towards using XFS for the SSD and EXT3 for the SATA.

My questions are: 1) Does this sound like a reasonable setup? 2) What filesystems would you recommend for the SSD and for the SATA?

Thanks

Jorisslob
  • 45
  • 1
  • 6
  • related: http://serverfault.com/questions/22565/advices-for-filesystem-choice-and-tuning-on-ssd – Martin M. Jun 10 '09 at 09:51
  • May I recommend you don't store files in a database. That way leads pain. – David Pashley Jun 10 '09 at 13:27
  • Ah, sorry for being unclear, what I meant is that the files will be sent to the database server. They will be stored on filesystem. Only the path is stored in the db. Thanks for suggesting it though. – Jorisslob Jun 10 '09 at 14:01

3 Answers3

2

On Solaris, ZFS has a feature that enable it to use SSD as cache devices.

But on Linux, i think you should use at least ext4 because it has some nice allocate-on-flush feature. Then, i would put all the large media on the SATA drive which has faster throughput than SSD (which is good for accessing lots of smalls files / random IO).

Anyway, where's your RAID mirror ?

Benoît
  • 1,331
  • 3
  • 11
  • 23
  • So even though they are large files, the SATA drive would be the best option to put those files? Actually, we will have a dual server. I'm not totally sure how that will work out (this is my first time to actually configure a server). – Jorisslob Jun 10 '09 at 10:11
  • SSD's are good for times when you won't be writing very often. – Brad Gilbert Jun 10 '09 at 16:21
  • Any modern (Sep 2013, but for a good year or so prior to now at least) SSD will massively outperform a SATA drive for both sequential and random throughput. Current consumer grade cheap SSDs can sustain 400MB/sec sequential access easily, whereas the best 7200RPM SATA drive will peak at around 180MB/sec on the outer edge of the drive and around half that on the inner edge. – Daniel Lawson Sep 18 '13 at 04:22
1

There's very little reason not to use ext3. It works very well in pretty much every situation. You may want to turn on dirindex if you have a lot of files in the same directory.

There are a number of special filesystems that take advantage of the way SSD drives work, but these aren't really stable yet. Maybe in another year, but that doesn't really help you now.

David Pashley
  • 23,151
  • 2
  • 41
  • 71
0

We've been examining the SSD issue too for performance.

What we have learned is:

1.) solid state disks can't be rewritten forever. Eventhough there isnt any mechanical arms nor motors, they will still have a maximum number of "writing" to their blocks = we need to make sure data has a failover somehow (have a wellfunctioning backup + fast restoring mechanism for the day that it happends)

2.) SSD because of large files? sounds like a very expensive solution, if the files are "that large". What are we talking about? 20MB? 100MB? 500 GB?

3.) Webserver, well - how fast is your outgoing connection then? are you sure that a RAID setup with normal "slow disks" isnt plenty? is it the access time pr. file you are concerned about?

I think if you use the SSD for a "fast cache" of loads of small files, its a great replacement of normal memory as its fast to find and stream from the SSD than buying loads of RAM and trying to fit enough into one server.

But if you have a lot of READ/WRITE access such as a large SQL-database, it seems that the SSD might die too soon.

Regarding filesystem, I am not the most experienced to tell you what to choose, but do you have any idea of the size or amount of files yet?

BerggreenDK
  • 188
  • 9
  • SSDs do have a limited number of write cycles, however this is very predictable and can be monitored over time. Also, if you're careful about SSD selection (and you don't need to speed thousands of dollars these days), you can get SSDS that won't practically wearout. – Daniel Lawson Sep 18 '13 at 04:26