We are looking to make a hardware purchase in the near-term and have a question:
Is there a necessity to go with SAS versions of SSD's? Or, are SATA versions sufficient?
Also, we are weighing the options of the all-flash and hybrid (SSD & HDD) storage models. Please provide your experience with either.
Thank you in advance for your response,
-Tom
Question
SAS vs SATA SSD Throughput
This topic has been closed for replies.
AFAIK all Nutanix certified hardware vendors rely on SAS for SSD and HDD.
What type and size of hardware you need entirely depends on the kind of workloads you are going to run on your cluster. Sizing a cluster correctly takes an experienced SE as mistakes made during this phase of the project can be very expensive to rectify later.
What type and size of hardware you need entirely depends on the kind of workloads you are going to run on your cluster. Sizing a cluster correctly takes an experienced SE as mistakes made during this phase of the project can be very expensive to rectify later.
MMSW_DE,
Thank you for your input. We have been working with a Nutanix SE but the quote we received did not specify what the underlying architecture was using.
I wanted to reach out to the community to see what their experience was with relation to the different storage models and/or interface types. Any other input is greatly appreciated.
-Tom
Thank you for your input. We have been working with a Nutanix SE but the quote we received did not specify what the underlying architecture was using.
I wanted to reach out to the community to see what their experience was with relation to the different storage models and/or interface types. Any other input is greatly appreciated.
-Tom
+4This question is very workload dependent. If this the workloads are largely heavy write workloads then all flash is your guaranteed performance, if this is generally more web servers then hybrid clusters should be fine. In personal experience when using hybrid clusters, go with the 3.84 TB ssds as your minimum 2 ssds. The 1.92 TB SSDS are fine but the 3.84s give you more room for burst workload if you are unsure about your workload and hybrid is the choice.
Given that the distributed storage fabric stages writes on the the SSDs first, having 10k + rpm spindle drives is not necessarily needed for initial writes. We have roughly 10k workloads in hybrid clusters (light app and web) and heavier io profiles, like databases are landed on all flash clusters. The idea is that the reads and writes are serviced out of the flash tier and the cold tier is only used when needed.
My advice would be to do your best to profile the IO om the machines you have today and use that as a basis for your decision.
Given how the DSF works you more than likely won't see any realized gain with SAS over SATA unless you really have hyper latency sensitive workloads and if that's the case then you should go with all flash or if you choose hybrid then elect to force those workloads to always reside on the ssds (flash pinning). Just be conscious of your other workloads if you do this.
Given that the distributed storage fabric stages writes on the the SSDs first, having 10k + rpm spindle drives is not necessarily needed for initial writes. We have roughly 10k workloads in hybrid clusters (light app and web) and heavier io profiles, like databases are landed on all flash clusters. The idea is that the reads and writes are serviced out of the flash tier and the cold tier is only used when needed.
My advice would be to do your best to profile the IO om the machines you have today and use that as a basis for your decision.
Given how the DSF works you more than likely won't see any realized gain with SAS over SATA unless you really have hyper latency sensitive workloads and if that's the case then you should go with all flash or if you choose hybrid then elect to force those workloads to always reside on the ssds (flash pinning). Just be conscious of your other workloads if you do this.
+6I agree with rhunt. Need to review performance (things like read/write percentages sequential or of VMs throughout a day, and plan for maximums (not avgs) - not to mention factoring in growth at the VM and cluster lever.
i.e. if you have MIXED workloads then I would recommend larger SSDs so your working sets don't end up in HDD land...
From what I could tell most HDDs are just plain SATA Enterprise drives (at least that's what G4's used to have.)
In another G5 cluster Some of the HDDs are ST8000NM0055-1RM112 which are definitely Enterprise SATA drives.
I would recommend talking to your SE - see if they have any resources to help measure/assist in this... don't do it alone.
i.e. if you have MIXED workloads then I would recommend larger SSDs so your working sets don't end up in HDD land...
From what I could tell most HDDs are just plain SATA Enterprise drives (at least that's what G4's used to have.)
In another G5 cluster Some of the HDDs are ST8000NM0055-1RM112 which are definitely Enterprise SATA drives.
I would recommend talking to your SE - see if they have any resources to help measure/assist in this... don't do it alone.
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.