0

This question already has an answer here:

We're a software development company and previously outsourced the HW side of business but now we're looking into building our own private cloud. We recently purchased a few servers, one of which is supposed to act as a central storage. The specs are as follows:

  • Chassis: CSE-826BE16-R920LPB Motherboard: Supermicro X10SLL-F CPU
  • E3-1200v3,Intel C222, 2GbE, Up to 32GB DDR3 ECC,2xSATA3,4xSATA2, 4xDDR3 IPMI
  • 1x CPU Intel Xeon E3-1220v3 - 3.1GHz, 8MB cache, 4core, HT, LGA1150, 80W
  • 1x heatsink SNK-0046A4 Activ 2U
  • 4x 8GB Samsung M391B1G73QH0-CK0 1600MHz DDR3 ECC Unbuffered 2R×8
  • 2x SSD 80GB Intel DC S3500 Series 2,5" SATA3, read 340 MB/s, write 100 MB/s
  • 10x HDD 2TB Seagate Constellation ES.3 ST2000NM0023 3,5" SAS2, 7200rpm, 128MB

The Seagate drives are set up as RAID 6 array. The SSDs are in RAID 1 and act as maxCache container which we turn on and off using maxView Storage Manager.

My first question - does this configuration make sense for a central storage of a private cloud where we plan to have other 3 compute nodes with two CPUs and lots of RAM on them?

My second question - would a similar configuration with less HDDs make sense for a MySQL database server for a reporting system with many concurrent requests? Or would it make more sense to use the SSDs for MySQL's temp space where it creates temporary tables.

And now for the main question... I tried to measure performance of this server with and without SSD cache (maxCache). The best tool I came across was iozone (iozone -a -g 8G) which produced these charts: http://www.bugweis.com/storage/comparison.zip. I am totally puzzled as it looks like the performance with maxCache is in most cases lower than without it.

I was wondering if iozone is a good way to test for real-life scenarios or if I'm doing this all wrong.

marked as duplicate by Iain, Dave M, Ward, Jenny D, sciurus Jan 31 '14 at 20:39

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

1

does this configuration make sense for a central storage of a private cloud where we plan to have other 3 compute nodes with two CPUs and lots of RAM on them?

I'd be worried that you have a single-point-of-failure here as it will only act as a single-controller storage box, you've also not really got a lot of memory (get more, it's cheap) and it looks like your only networking is 2 x 1Gbps NICs - they'll be a massive choke-point in this setup. So no I wouldn't recommend this - what price have you been given for this?

would a similar configuration with less HDDs make sense for a MySQL database server for a reporting system with many concurrent requests? Or would it make more sense to use the SSDs for MySQL's temp space where it creates temporary tables.

I'd be tempted to go for a two socket system for a DB box, even if you don't use the second socket right away - database engines tend to scale well with more cores. Again you'd want a lot more memory. As for what disks to use - well it would be foolish to make any suggestions here as DB IO requirements vary more than just about any part of an infrastructure based on the read/write balance and a lot of other factors such as the table sizes etc.

  • Yeah the single point of failure is definitely a concern but at this point we cannot afford to build a highly available infrastructure. You said 32G is not a lot of RAM - well how much would you suggest for a server that only works as a data storage? We paid about 3800 EUR for this server. – Libor Suchý Jan 31 '14 at 12:53
  • The network cards both have two physical ports so it's actually 4 x 1Gbps. Still a potential bottleneck I presume? – Libor Suchý Jan 31 '14 at 12:56
  • Sorry, just realized the description is missing the number of RAM units. – Libor Suchý Jan 31 '14 at 12:57
  • I'd still rather use a dual-controller system such as a HP P2000 G3 array – Chopper3 Jan 31 '14 at 13:17
0

UPDATE - this is what Adaptec support said:

The benefit of Maxcache writes and reads occurs only when the system has run for a long enough period to populate the "Hot Data" on the ssd drives. This would take some time depending on the data written and the requests made to the system, so a short benchmarking test may not show any improvement, as it may not allow enough time for the "Hot data" to populate the drives.Please also make sure that the outstanding I/O values are set 32 or 64, as some benchmarking software packages set the deafult value to 1, whcih would affect performance values obtained and would not be correct.

So I guess iozone is not the best way to test maxCache performance.

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