F5F Stay Refreshed Power Users Networks High-speed 10 Gbit network paired with an SSD storage solution ideal for scientific applications.

High-speed 10 Gbit network paired with an SSD storage solution ideal for scientific applications.

High-speed 10 Gbit network paired with an SSD storage solution ideal for scientific applications.

M
MonskristCreed
Junior Member
4
08-04-2017, 12:09 PM
#1
Hello everyone, I’m a PhD student specializing in structural biology and we’re currently organizing our lab’s computing setup. Our IT support here is limited, so we’re handling everything ourselves. I’d like your assistance with this. First, I want to briefly outline our requirements: We’re working with substantial data sets—roughly 150 to 500 GB each, containing around 5,000 to 20,000 files (each about 32 MB). These need to be loaded, processed, and written repeatedly using Matlab. The work can be efficiently parallelized, leveraging both CPU and GPU resources (CUDA). We currently have four identical workstations running in a Matlab cluster to handle one dataset at a time. This process often takes up to a week. To run them simultaneously, the files must reside on a central server where all stations can access them together. Presently, we rely on a 1 Gbit network, and loading or saving processed files is currently the slowest part. That’s why we’re considering installing a small local server and linking all workstations to it via a 10 Gbit connection. All stations already have 10 Gbit network cards. Our available hardware includes ten Intel CNA X710-DA2 cards with SFP+ direct connections. Here are the specifications of each station: Mainboard: Gigabyte X299 UD4-CF CPU: i9-7980XE GPU: Nvidia GTX 1080 Ti RAM: 128 GB Crucial CT16G4DFD824A Network: Intel CNA X710-DA2 Power Supply: Supermicro PWS-903-PQ I’m not familiar with 10 Gbit networking, but any recommendations for a suitable switch and server would be greatly appreciated. Ideally, we’d aim for around 2 TB of storage—perhaps eight 512 GB SSDs in RAID10—to accommodate the active project on the server. We have some budget available, but we need it to be reasonable. Thank you all for your help!
M
MonskristCreed
08-04-2017, 12:09 PM #1

Hello everyone, I’m a PhD student specializing in structural biology and we’re currently organizing our lab’s computing setup. Our IT support here is limited, so we’re handling everything ourselves. I’d like your assistance with this. First, I want to briefly outline our requirements: We’re working with substantial data sets—roughly 150 to 500 GB each, containing around 5,000 to 20,000 files (each about 32 MB). These need to be loaded, processed, and written repeatedly using Matlab. The work can be efficiently parallelized, leveraging both CPU and GPU resources (CUDA). We currently have four identical workstations running in a Matlab cluster to handle one dataset at a time. This process often takes up to a week. To run them simultaneously, the files must reside on a central server where all stations can access them together. Presently, we rely on a 1 Gbit network, and loading or saving processed files is currently the slowest part. That’s why we’re considering installing a small local server and linking all workstations to it via a 10 Gbit connection. All stations already have 10 Gbit network cards. Our available hardware includes ten Intel CNA X710-DA2 cards with SFP+ direct connections. Here are the specifications of each station: Mainboard: Gigabyte X299 UD4-CF CPU: i9-7980XE GPU: Nvidia GTX 1080 Ti RAM: 128 GB Crucial CT16G4DFD824A Network: Intel CNA X710-DA2 Power Supply: Supermicro PWS-903-PQ I’m not familiar with 10 Gbit networking, but any recommendations for a suitable switch and server would be greatly appreciated. Ideally, we’d aim for around 2 TB of storage—perhaps eight 512 GB SSDs in RAID10—to accommodate the active project on the server. We have some budget available, but we need it to be reasonable. Thank you all for your help!

D
Depths_
Junior Member
48
08-04-2017, 12:48 PM
#2
I'm not confident about the 10GBit/s speed because the files are tiny and your window won't expand much before they're sent... At least for a network file share, I think iSCSI could work better.
D
Depths_
08-04-2017, 12:48 PM #2

I'm not confident about the 10GBit/s speed because the files are tiny and your window won't expand much before they're sent... At least for a network file share, I think iSCSI could work better.

I
ItzCh3nTi_YT
Member
96
08-05-2017, 11:33 AM
#3
No, there is no such data available.
I
ItzCh3nTi_YT
08-05-2017, 11:33 AM #3

No, there is no such data available.

O
Outrider2K
Junior Member
42
08-06-2017, 11:48 PM
#4
Delay is minimal for quick data bursts, particularly in render farms or folding systems.
O
Outrider2K
08-06-2017, 11:48 PM #4

Delay is minimal for quick data bursts, particularly in render farms or folding systems.

M
Matke04
Posting Freak
825
08-12-2017, 01:14 PM
#5
I'll discuss 10Gb switch suggestions later. For many smaller files, SMB Direct makes a big difference. Just need RDMA support—Intel X710s have that, and ensure your file server does too. https://docs.microsoft.com/en-us/windows...smb-direct SAMBA also offers SMB Direct Support if you're on Linux, which seems like the first question I should have asked. Which operating system are you running?
M
Matke04
08-12-2017, 01:14 PM #5

I'll discuss 10Gb switch suggestions later. For many smaller files, SMB Direct makes a big difference. Just need RDMA support—Intel X710s have that, and ensure your file server does too. https://docs.microsoft.com/en-us/windows...smb-direct SAMBA also offers SMB Direct Support if you're on Linux, which seems like the first question I should have asked. Which operating system are you running?

S
SoyDash
Posting Freak
859
08-24-2017, 05:57 AM
#6
Hey, sorry for the late response. Thanks a lot for all your answers. Regarding the small file sizes, I am not sure it matters that much, because in the workflow usually a large portion of the dataset (let's say 1000 x 32 mb file, so roughly 32GB) is loaded at the same time and then processed. We are using centOS7 and the network server is connected by SMB. The server we are connected to is a "Dell EMC Isilon". I will look into SMB Direct and iSCSI. @leadeater It would be really great if you could recommend an 10Gb switch. Thanks a lot Steffen
S
SoyDash
08-24-2017, 05:57 AM #6

Hey, sorry for the late response. Thanks a lot for all your answers. Regarding the small file sizes, I am not sure it matters that much, because in the workflow usually a large portion of the dataset (let's say 1000 x 32 mb file, so roughly 32GB) is loaded at the same time and then processed. We are using centOS7 and the network server is connected by SMB. The server we are connected to is a "Dell EMC Isilon". I will look into SMB Direct and iSCSI. @leadeater It would be really great if you could recommend an 10Gb switch. Thanks a lot Steffen

C
Cupcake_Rose
Posting Freak
844
08-25-2017, 03:23 PM
#7
Our pricing covers a variety of options to suit different needs.
C
Cupcake_Rose
08-25-2017, 03:23 PM #7

Our pricing covers a variety of options to suit different needs.

S
S3xyOG
Junior Member
17
09-05-2017, 07:25 PM
#8
S
S3xyOG
09-05-2017, 07:25 PM #8