Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I would like to know, how to calculate sustained read and write speeds supported. I don't have access to machine/drive, but need to define specs to purchase one.

What all things would it be dependent on?

Using RAID 10, would improve it?

Would it be dependent on file size, which we read or write?

How to calculate latency of fetching a file segment? Seek time etc..

I need to use it with a web server, for storing and delivering large no. of small files. There will be continuous writes and reads 24x7 of small files. Any recommendations for which RAID to use, or no RAID should be used, based on writes or reads?

share|improve this question

A call to Dell should give you all of the information that you need. There are white papers and tech specs on all of this available if you call your rep.

Raid 10 is pretty much universally used where performance matters, but you should always test multiple scenarios. You could always be an edge case.

share|improve this answer
I need to get practical numbers, from someone who has used it. Also what all things would effect performance etc. – Nagota Sep 30 '11 at 17:04
@Nagota No one has your setup, so no one can benchmark for it, especially with the sparse information that you've provided. Dell (and other vendors) provide practical numbers. If they spec a drive for a certain set of performance metrics and it doesn't mean them, you'd better believe that you can have that drive replaced for operating out of spec. – MDMarra Sep 30 '11 at 17:08

Each 7.2k RPM drive can perform between 80-150 IO/s, assuming the workload is random small reads and writes. Writes will usually be cached by the raid controller, so will go as fast as the memory does until it's full. Raid 10 has a minimum of 4 drives, so your reads will be between 320 and 600 IO/s with the minimum number of drives.

If you choose raid 5, you will have more usable space, but your IO/s per drive on average might go down, depending on the raid controller you have and what percent of your workload is writes.

IO/s is a measure of performance that depends heavily on seek time, and an average seek time of about 6 ms is average for 7.2k RPM drives. You can get better by upgrading to 10 or 15k RPM, but you could also increase your effective read "speed" by increasing the number of 7.2k drives you stripe your data across.


  • Size your workload performance, and then fit your drives to meet or exceed it. If you think your web server is capable of doing 500 IO/s, go for 6 7.2k RPM drives.
  • Choose your raid carefully- if you don't need space as much as you need speed, go with raid 10. It also allows you to lose as many as half your drives before having a loss of data. Raid 6 is also popular for large drives as it can lose two drives before failing, however introduces a heavy workload on your raid card.
  • Invest in a decent battery backed cache raid card.
share|improve this answer

MarkM already said the important thing here. All I can add to that is that you need to keep in mind the fact that NL-SAS is really a simple SATA drive with a SAS interface, and if you want better performance, you need to look at real SAS drives. In any case, Dell is the first and foremost address for information about their equipment, nobody can know it better than them

share|improve this answer
You can always make up for lower performance per spindle by putting more spindles onto the job. That said, if space really isn't a matter at all, then SAS drives can be cheaper than large 7.2k drives, where you're paying for capacity. – Basil Oct 3 '11 at 15:42
all true and agreed upon, but not really relevant to my comment :) – dyasny Oct 3 '11 at 15:52

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.