Mini cluster specifications for sequence alignement

Won-Sul · 04-26-2016, 05:11 AM

Hello everybody!

I am a postdoc in a brand new small research group, and I am in charge of setting the the IT structure for analysis. My boss would like to buy a mini cluster (budget around 10 - 30k euros), but I have no idea what specs I should look for. I'd be interested in having your opinion. I was thinking of a machine with maybe 4 nodes (we are a small group at the moment).

We are going to do a lot of whole genome bisulfite sequencing mapping, using plant genomes (mostly Arabidopsis, but also maybe tomatoes or other plants with a larger genome). ChIP-seq mapping and SNP calling over large amount of samples will also be done. We don't plan to do genome assembly, but it may happen in the future.

I thank you all in advance for your advice

Sincerely

Markiyan · 04-29-2016, 05:53 AM

Dear Won-Sul,

With this type of budget one is likely to be looking at supermicro/dell based equipment.
Make sure you find some good local company for assembling your server(s), and have some room for expansion in the future and lots of DAS (direct attached storage in RAID10)

If getting rackmount equipment - make sure you have somewhere uninhabited to put it, (90db+ noise 24/7), and make sure you use 4U+ cases for a headnode and 2U+ for a worknodes. You can get only one worknode initially, and see how it works in a real life (Quite often it is better to have one 256GB box than a bunch of 64GB boxes) - esp for de novo type of work.

One year ago we've got ourself something like this:
Chassis: Broadberry 4U 24 Hot-Swap Bay with Redundant 920Watt Efficient Power Supply, 12Gb/s
Backplane
Motherboard X10DRi-T, 16 Dimm Slots (upto 1TB RAM), Dual Intel 10GbE LAN, On Board Graphics, On
Board SATA RAID 0,1, IPMI & Remote KVM
Processor 2x E5-2667v3 Intel 8 Core Xeon 3.20GHz 25Mb Cache 9.6 FSB
Memory 8x 32GB 2133MHz DDR4 ECC Registered DIMM Module
+ HW Raid controller (LSI or Adaptec).
And then put in 10 4TB SAS HDD's and 4x 1TB SATA3 SSD's for system/home/blastdb (Samsung 850 Pro, Queue depth=1 for SSD's).
Also did some case modding, so the sata3 SSD's are connected directly to the motherboard's SATA3 ports, and added a couple of fans for better cooling of the network card/etc heatsinks.

If you don't have dedicated equipment room and need it to be quiet - than the Lian Li PC-D8000B 'Feel the Space' Full Tower Case would be way better choice for a headnode, coupled with two Corsair
Hydro H110 280mm High Performance Liquid CPU Coolers (CW-9060014-WW).
(I have one of them on my i7 based workstation).
You can put up to 20 3.5' HDD's in there, and it has vibration damping rubber gourmets.

Thecus N7700Pro makes more noise than that Lian Li Beast Case with 7x12cm + 2x14cm fans, while happily churning out 400-500W of heat under full blast load.

Won-Sul · 05-02-2016, 07:43 AM

Hi Markiyan

Thank you very much for your extremely detailed answer. I really appreciate!

Cheers

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
MiSeqDx specifications	alunem	Illumina/Solexa	4	02-10-2014 10:13 AM
calculate alignement differences	atma_weapon	Bioinformatics	0	04-11-2013 04:22 PM
local mode alignement of reads	StephaniePi83	Bioinformatics	0	06-07-2012 08:18 AM
bwa alignement outside of the reference sequence	cklopp	Bioinformatics	1	09-14-2011 09:25 AM
Questions about hardware specifications for analysis	sallusti10	Illumina/Solexa	7	02-09-2010 12:39 PM

04-26-2016, 05:11 AM	#1
Won-Sul Junior Member Location: France Join Date: Feb 2012 Posts: 2	Mini cluster specifications for sequence alignement Hello everybody! I am a postdoc in a brand new small research group, and I am in charge of setting the the IT structure for analysis. My boss would like to buy a mini cluster (budget around 10 - 30k euros), but I have no idea what specs I should look for. I'd be interested in having your opinion. I was thinking of a machine with maybe 4 nodes (we are a small group at the moment). We are going to do a lot of whole genome bisulfite sequencing mapping, using plant genomes (mostly Arabidopsis, but also maybe tomatoes or other plants with a larger genome). ChIP-seq mapping and SNP calling over large amount of samples will also be done. We don't plan to do genome assembly, but it may happen in the future. I thank you all in advance for your advice Sincerely

04-29-2016, 05:53 AM	#2
Markiyan Senior Member Location: Cambridge Join Date: Sep 2010 Posts: 116	Get yourself some good headnode, and add worknodes if needed. Dear Won-Sul, With this type of budget one is likely to be looking at supermicro/dell based equipment. Make sure you find some good local company for assembling your server(s), and have some room for expansion in the future and lots of DAS (direct attached storage in RAID10) If getting rackmount equipment - make sure you have somewhere uninhabited to put it, (90db+ noise 24/7), and make sure you use 4U+ cases for a headnode and 2U+ for a worknodes. You can get only one worknode initially, and see how it works in a real life (Quite often it is better to have one 256GB box than a bunch of 64GB boxes) - esp for de novo type of work. One year ago we've got ourself something like this: Chassis: Broadberry 4U 24 Hot-Swap Bay with Redundant 920Watt Efficient Power Supply, 12Gb/s Backplane Motherboard X10DRi-T, 16 Dimm Slots (upto 1TB RAM), Dual Intel 10GbE LAN, On Board Graphics, On Board SATA RAID 0,1, IPMI & Remote KVM Processor 2x E5-2667v3 Intel 8 Core Xeon 3.20GHz 25Mb Cache 9.6 FSB Memory 8x 32GB 2133MHz DDR4 ECC Registered DIMM Module + HW Raid controller (LSI or Adaptec). And then put in 10 4TB SAS HDD's and 4x 1TB SATA3 SSD's for system/home/blastdb (Samsung 850 Pro, Queue depth=1 for SSD's). Also did some case modding, so the sata3 SSD's are connected directly to the motherboard's SATA3 ports, and added a couple of fans for better cooling of the network card/etc heatsinks. If you don't have dedicated equipment room and need it to be quiet - than the Lian Li PC-D8000B 'Feel the Space' Full Tower Case would be way better choice for a headnode, coupled with two Corsair Hydro H110 280mm High Performance Liquid CPU Coolers (CW-9060014-WW). (I have one of them on my i7 based workstation). You can put up to 20 3.5' HDD's in there, and it has vibration damping rubber gourmets. Thecus N7700Pro makes more noise than that Lian Li Beast Case with 7x12cm + 2x14cm fans, while happily churning out 400-500W of heat under full blast load.

05-02-2016, 07:43 AM	#3
Won-Sul Junior Member Location: France Join Date: Feb 2012 Posts: 2	Hi Markiyan Thank you very much for your extremely detailed answer. I really appreciate! Cheers