Common Workloads That Should Never Run on Shared Storage
-
So, broken down, it does no good to have two or three DCs for failover but have them on the same datastore, even though they'd obviously be separate VMDKs.
-
@thanksaj said:
So, broken down, it does no good to have two or three DCs for failover but have them on the same datastore, even though they'd obviously be separate VMDKs.
Right?
-
@thanksaj said:
So, broken down, it does no good to have two or three DCs for failover but have them on the same datastore, even though they'd obviously be separate VMDKs.
Correct, they are already replicating and completely independent of each other. Taking those already completely replicated systems and tricking them into thinking they are separate but actually putting them all together on a single storage device de-replicates them at the storage layer defeating the purpose.
It's the equivalent to taking a backup and keeping the backup on the same disk as the original data.
-
@scottalanmiller said:
@thanksaj said:
So, broken down, it does no good to have two or three DCs for failover but have them on the same datastore, even though they'd obviously be separate VMDKs.
Correct, they are already replicating and completely independent of each other. Taking those already completely replicated systems and tricking them into thinking they are separate but actually putting them all together on a single storage device de-replicates them at the storage layer defeating the purpose.
It's the equivalent to taking a backup and keeping the backup on the same disk as the original data.
Ok, yeah. This makes total sense. In reality, every business should have a minimum of two physical servers, with the replicating VMs on separate physical boxes.
-
@thanksaj said:
In reality, every business should have a minimum of two physical servers, with the replicating VMs on separate physical boxes.
Not all all. Many small businesses, easily most, have no business value to having a second server. Needing any type of failover is something that must be justified through lost profits caused by downtime. For the typical SMB, downtime of a day or two is not enough to justify the cost of a second server. You need to be of a certain size coupled with a profit dependency on your server-based workloads to justify just having secondary hardware, let alone having a replication strategy.
The most common architecture for any workload is the single server, no replication. Also known as the "mainframe architecture."
-
@scottalanmiller said:
@thanksaj said:
In reality, every business should have a minimum of two physical servers, with the replicating VMs on separate physical boxes.
Not all all. Many small businesses, easily most, have no business value to having a second server. Needing any type of failover is something that must be justified through lost profits caused by downtime. For the typical SMB, downtime of a day or two is not enough to justify the cost of a second server. You need to be of a certain size coupled with a profit dependency on your server-based workloads to justify just having secondary hardware, let alone having a replication strategy.
The most common architecture for any workload is the single server, no replication. Also known as the "mainframe architecture."
I guess, then, it's one solid server and a separate backup strategy. Maybe that means running the backup VM (like a UEB) on the same server but archiving to a NAS or the like. However, that would still be a huge risk in and of itself.
-
Totally disagree. I think most businesses should have two physical servers.
-
@Carnival-Boy said:
Totally disagree. I think most businesses should have two physical servers.
Regardless of its suitability or financial justification? What factor makes you feel that they should need two even when financially they can't justify the value?
-
Woah there! That's a leading question. I don't operate in a bubble you know, all my opinions take into account the financial aspects.
For very small businesses, say less than 50 users, I can imagine single servers being the norm. These are companies that have traditionally run everything off a single install of SBS or whatever the equivalent now is. This isn't my area though.
But above 50 users, or what I'd call medium sized, then I think most would want/need at least two servers. Of course, every case is different.
-
@Carnival-Boy said:
For very small businesses, say less than 50 users, I can imagine single servers being the norm. These are companies that have traditionally run everything off a single install of SBS or whatever the equivalent now is. This isn't my area though.
This is the biggest pool of businesses, though. At least in the US the vast majority of businesses are less than 50 users, by far.
-
@Carnival-Boy said:
But above 50 users, or what I'd call medium sized, then I think most would want/need at least two servers. Of course, every case is different.
Number of users is a decent rule of thumb indicator although it isn't as useful as it seems. Things like factories often have near zero computer dependence on day to day operations but have a high number of workers (this is non-enterprise factories, of course) whereas a small accounting firm has few people but a high computer dependence. So in one case, 150 or more people might work find from a single server whereas in the other only thirty, perhaps.
It is amazing, though, how much a reliable failover system costs (nearly double not having one) and equally amazing how often downtime has very little financial impact on a small business.
-
I have three servers, though two would be fine. I don't have what I understand as a "reliable failover system" or anything close to it, but if one server is down I am in a position to provide a degraded or limited service using the other server(s) which is generally enough to keep users happy and the business ticking over. It's in no way automated or anything like high availability, but it gives me options and in a crisis I like to have as many options as I can. The cost isn't anywhere near double, since you're not doubling up on disks or memory or CPU by spreading the load across two boxes.
