Cost Study: 4 Node Scale vs. 4 Node VMware IPOD
-
@Dashrender said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@Dashrender said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
Scott didn't mention the RAID controller in his Dells, what kind of RAID are in your boxes? how about the warranty, what are the details?
No RAID with hyperconverged players. Not that there couldn't be, but no one uses one. Since everyone has to have RAIN anyway at a higher layer, having a RAID controller doesn't add any value. So none in the Scale, VSAN, Starwind or IPOD examples. Not just no RAID controller, no RAID at all. HTBase is the same in that regard.
for the uninitiated, you just blew my mind a little.
@John-Nicholson and I have been talking about the death of RAID for years. RAID pretty much exists as a vestige for very small environments that still see their infrastructure in terms of "a single server" and not as clusters and clouds. Once you get beyond the "each node handles its own storage" point (which only applies to one or possibly two host clusters) RAID has no value. Gluster, CEPH, and anything perceived as "cloud storage" and anything like VSAN, Starwind or hyperconvergence are all RAID-less. We've long been in the post-RAID world, RAID remains almost solely for the smallest SMBs.
-
We were on the cusp of being post-RAID at the Wall St. shop I was in by 2006. We were evaluating non-RAID clustering but it just wasn't there in performance terms at the time. We were able to use RAID to go faster for our big performance needs, but that was a long time ago. Luster was a major push to change that.
-
@Dashrender Yep, thats right!
-
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
We were on the cusp of being post-RAID at the Wall St. shop I was in by 2006. We were evaluating non-RAID clustering but it just wasn't there in performance terms at the time. We were able to use RAID to go faster for our big performance needs, but that was a long time ago. Luster was a major push to change that.
What are alternatives to Lustre today? Gluster?
-
@dafyre HTFS from HTBase too!
-
@htbase said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@dafyre HTFS from HTBase too!
Is the HTFS free or open source?
If it's not free, how is it priced -- per host, per TB... etc?
-
@dafyre said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@htbase said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@dafyre HTFS from HTBase too!
Is the HTFS free or open source?
If it's not free, how is it priced -- per host, per TB... etc?
And can it be used independently? What interfaces are available for it?
-
@dafyre Its not free or open source right now. Its charged on a host/node base
Yep, it can be used independently from our hyperconverged solution
-
@htbase said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@dafyre Its not free or open source right now. Its charged on a host/node base
Yep, it can be used independently from our hyperconverged solution
Thanks for the heads up!
Can you share list pricing per host?
-
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
A cost difference of $67,594 is pretty significant for any SMB.
Because you picked something I would buy, well, not me because I would run out of space first on it, not an SMB. Dell still sells Compellent storage. An SCv2020, same amount of disks, with support, is $30K on Fibre Channel. Why not quote PURE SANs while we are at it, stacked to the gills with the fastest SSDs we can find?
Cherry picking storage is always gonna make hyperconvergence look good. You need to review your workload to determine if that's gonna be the right path. Most SMBs workload only needs cheap and deep storage because they are all digital packrats and very little compute. Most SMBs workload would be better ran in cloud services. So really, it's not about cost study for local servers, it's about if it's better to host outside and get a giant NAS for those local files which people can't seem to get rid of.
-
@PSX_Defector said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
A cost difference of $67,594 is pretty significant for any SMB.
Because you picked something I would buy, well, not me because I would run out of space first on it, not an SMB. Dell still sells Compellent storage. An SCv2020, same amount of disks, with support, is $30K on Fibre Channel. Why not quote PURE SANs while we are at it, stacked to the gills with the fastest SSDs we can find?
Cherry picking storage is always gonna make hyperconvergence look good. You need to review your workload to determine if that's gonna be the right path. Most SMBs workload only needs cheap and deep storage because they are all digital packrats and very little compute. Most SMBs workload would be better ran in cloud services. So really, it's not about cost study for local servers, it's about if it's better to host outside and get a giant NAS for those local files which people can't seem to get rid of.
Not cherry picked. Literally the closest competitor with their most aligned product. It's a product that is significantly below the standard of the HC solution and the minimum we could possibly consider to be competition. Going to Pure is ridiculous because it is so much more and not comparable. Compellent is "sold" but is lower end than is seriously consideration in apples to apples and as it is now at risk for support, it's not considered a viable competitor anyway. Cherry picking is the opposite of what was done here, this is as not cherry picked as we can get.
Find 3PAR or HDS pricing for hybrid storage if you want other options in a reasonable range. Anything "less" is cherry picking to lower the price while changing the goals.
-
@PSX_Defector said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
Because you picked something I would buy...
I thought you worked for a serious hosting company. That's entry level SMB storage. VNXe is super low end highly reliable storage.
-
And here shows the constant issue with finding a solution.
I think Scott is on the right path, trying to make the apples to apples comparison. I don't know anything personally about Compellent, so I'm not sure why Scott doesn't like it - he explained some, but not enough to overcome my lack of knowledge.
-
@Dashrender said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
And here shows the constant issue with finding a solution.
I think Scott is on the right path, trying to make the apples to apples comparison. I don't know anything personally about Compellent, so I'm not sure why Scott doesn't like it - he explained some, but not enough to overcome my lack of knowledge.
Compellent isn't bad, or wasn't. But it is a known "dead end", it's part of the Dell storage family that EMC is replacing. Right now, the market is considering any Compellent sale to be "dumping of old stock." The VNXe is what Dell bought to be the Compellent replacement. I don't know if Compellent has a product up to par with the VNXe, but they had good stuff. But the prices are not that much cheaper, and if they are in any way not on par with the VNXe, then they are "cutting corners" even more to make SAN look viable where it clearly is not.
As it is, the VNXe is "cutting corners" on speed, reliability and flexibility compared to the alternative solution to make the IPOD / SAN approach look better than we should, it's just what I feel is the "best" comparison as it's "good enough" to be a viable SAN while having specs that most closely match the other specs and is from the market leader. But even EMC only sells the VNXe as a five nines (e.g. not high availability) and entry level product, so it is not actually on toe to toe level with the six nines hyperconverged product that we are comparing against. We get five nines from a normal server, so the VNXe doesn't have the "order of magnitude" improvement that you should be getting from an HA solution.
-
Prices on Compellent are rumoured to be exceptionally good right now, for obvious reasons. Making it an invalid comparison point (dumping stock would only be viable if we were comparing against an older hyperconvergence solution that was also temporarily dumping stock before being discontinued.) It would be like using a fire sale to determine market pressures. It's just misleading.
The viable products here are players like EMC, HDS, Nimble and HPE 3PAR. Products good "enough" to qualify as HA in an IPOD design (even if not the same level of HA as we have with the HC design), and from current supported products that are still being developed and that are available in the sizes, speeds and styles (hybrid, iSCSI) that we need for the comparison. And, obviously, it has to come from a top tier player with the same level of support logistical infrastructure as someone like Dell or HPE for servers.
-
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
We get five nines from a normal server, so the VNXe doesn't have the "order of magnitude" improvement that you should be getting from an HA solution.
Did I miss a part where 6 nines was the goal? or HA was the goal?
-
@Dashrender said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
We get five nines from a normal server, so the VNXe doesn't have the "order of magnitude" improvement that you should be getting from an HA solution.
Did I miss a part where 6 nines was the goal? or HA was the goal?
That's a core focus of a cluster in general. If you were to not be focused on HA everything would change. You would need fewer nodes, no failover licensing, etc. While there is no guarantee that HC means HA, all HC clusters made today focus on HA as a core feature. A major feature. And unless you specifically over-provision your cluster, HA is a natural byproduct of the safer architecture. So to do apples to apples you always have to consider HA because it's just one of the components of the original design.
-
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@Dashrender said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
We get five nines from a normal server, so the VNXe doesn't have the "order of magnitude" improvement that you should be getting from an HA solution.
Did I miss a part where 6 nines was the goal? or HA was the goal?
That's a core focus of a cluster in general. If you were to not be focused on HA everything would change. You would need fewer nodes, no failover licensing, etc. While there is no guarantee that HC means HA, all HC clusters made today focus on HA as a core feature. A major feature. And unless you specifically over-provision your cluster, HA is a natural byproduct of the safer architecture. So to do apples to apples you always have to consider HA because it's just one of the components of the original design.
OK great - thanks for adding that to the post
-
Removing HA would also be complex because if we kept a high quality SAN (or NAS) the cost difference would go crazy towards the HC cluster (the IPOD cost would go down by almost nothing while the HC would drop by 25%) or if we went with a commodity SAN (like Synology or ReadyNAS) the reliability of the IPOD would plummet so precipitously that we wouldn't be apples to apples at all.
-
@scottalanmiller said in Cost Study: 4 Node Scale vs. 4 Node VMware IPOD:
@John-Nicholson and I have been talking about the death of RAID for years. RAID pretty much exists as a vestige for very small environments that still see their infrastructure in terms of "a single server" and not as clusters and clouds. Once you get beyond the "each node handles its own storage" point (which only applies to one or possibly two host clusters) RAID has no value. Gluster, CEPH, and anything perceived as "cloud storage" and anything like VSAN, Starwind or hyperconvergence are all RAID-less. We've long been in the post-RAID world, RAID remains almost solely for the smallest SMBs.
We have a Starwind cluster of two all-flash nodes that runs on top of hardware RAID5 making a redundancy over redundancy like RAIN1 on top of RAID5 which is quite awesome since there is a consistent set of data on each host in the cluster which is impossible with RAIN stuff like VMware VSAN or S2D does. I treat it like additional hardware offload for storage managing and it performs better than pure software RAID for sure.
