How many spares for C6100 to cover 1 year ?.

[RimBlock]

I am trying to hedge against a years warranty for a client on a large order of these but have no real idea on failure / spares level in order to offer that warranty.

1 extra for every 10, 1 for every 20 in order to cover the warranty ?.

Would be interested to hear peoples views.

RB

 

[korban]

Take a look at google stat Google spotlights data center inner workings | News Blogs - CNET News and Google?s Disk Failure Experience

They are using consumer component and not server parts, then it's important to extrapolate to C6100 coponent.

For me it depend of the componenet types.

In a 24/7 install I think you will need to replace often parts with capacitor (Motherboard, psu, ...) than "flat component" (cpu, ram,...)

Then, don't mind in term of C6100 bundle but separed pieces. It could be interesting to contact Peter Neiman from Servers, Dell Servers, Used Servers | The Server Store to check if he can sell you spare parts.
if it is not possible check here The Server Store and try to build a 2 node C6100 with 2 power supplie and check the cost/spare.


In global I will do that:

First 10 order I will take 20% then each extra 10%

 

[Andreas]

The economic model of used C6100 with their different position towards end of life and cheaper availability of whole systems influences the cost calculation.
1) It is reasonable to assume that used C6100's will continue to be available for the next 12-24 months, with potentially declining acquisition cost.
2) The probabilty curve for warranty incidents has its peak at the beginning of a cycle
3) Will you supply these servers only to one customer, or do you intend to broaden your C6100 business to multiple customers - spreading the probability further
4) You might need less coverage by static things like case, vs. faster worn things like PSU, drives.
5) You might overprovision on site, and agree with the customer a flexible warranty and/or payment model if they need to tap into the spare capacity on site. Reducing your labor cost in a warranty case.

Andy

 

[Chuckleb]

I feel that for the most part, these things don't fail. I have racks of compute clusters and the most often failed part is the HDD. I have one spare chassis (no memory or CPUs) with 4 sleds for the 10 that I have. Another group bout a similar spare shell for their 20 odd units. This should last us easily for the next 2-3 years, basically the useful life of the gear. I would not bother populating the spare parts.

 

[Patrick]

The economic model of used C6100 with their different position towards end of life and cheaper availability of whole systems influences the cost calculation.
1) It is reasonable to assume that used C6100's will continue to be available for the next 12-24 months, with potentially declining acquisition cost.
2) The probabilty curve for warranty incidents has its peak at the beginning of a cycle
3) Will you supply these servers only to one customer, or do you intend to broaden your C6100 business to multiple customers - spreading the probability further
4) You might need less coverage by static things like case, vs. faster worn things like PSU, drives.
5) You might overprovision on site, and agree with the customer a flexible warranty and/or payment model if they need to tap into the spare capacity on site. Reducing your labor cost in a warranty case.

Andy

Just as a thought: each spare chassis at ~$250 is probably going to cost you less than even buying one motherboard separately so might be worth it to get an extra. As Andy and others have mentioned, for 12-24 months you should be relatively OK but in terms of labor just having a cold spare chassis around at each site may be the easiest thing to do.

 

[dba]

I am trying to hedge against a years warranty for a client on a large order of these but have no real idea on failure / spares level in order to offer that warranty.

1 extra for every 10, 1 for every 20 in order to cover the warranty ?.

Would be interested to hear peoples views.

RB

Good question! I have no idea, really, but I do have an acecdote: Sun once paid me a few hundred dollars to be part of a customer panel to discuss expensive Sun 24x7 on site warranty services. I'm sure they wanted to learn how to sell us more such warranties, but it didn't work out. The consensus was that the cheapest solution for the customer was to buy an extra server or two to act as spares. I remember that one institution kept two spares for around 50 servers. Their "warranty cost" was therefore 4%, which was dramatically cheaper that Sun wanted to charge! In fact, buying one spare for two servers was cheaper.

 

[Patrick]

Exactly. There is a full spare C6100 chassis with 2x power supplies and 2x motherboard trays in the colo for 2x servers.

 

[PigLover]

RB,

Your question is very difficult to answer in the abstract. The 'right' number of spares depends on many variables that are specific to your application and expectations. Do you already have redundancy or resiliency built in to your application? Do you have spare capacity to cover an outage? For how long? Given your application, what types of faults create a service disruption? What does it take to recover that disruption? What is your expected service level and/or outage tolerance? What are your expectations about availability, delivery time or cost of the replacement nodes if you had to acquire it post-fault?

You can still figure out some generalities... The C6100 is exceptionally cheap and highly available right now. If you need fast/easy recovery then buying a spare chassis is a good idea (something like Patrick did for his colo). It's unlikely that you need more than 1 chassis per 10-20 installed. Your not going to get a better answer than this without some details about your application, architecture, service levels and risk tolerance.

I can tell you that in my business we have very tight service level expectations ("5 nines" or better) and no tolerance for outages. We generally budget 10% of hardware costs for on-site spares, but we also run with both local and geographic hot redundancy for all of our systems and in many cases our architecture allows us to run our peak loads with entire sites offline (we keep lots of unused capacity on line). Without the redundancy and service margin our 'spares' requirement would be man times higher than it is.

 

[RimBlock]

Your question is very difficult to answer in the abstract. The 'right' number of spares depends on many variables that are specific to your application and expectations.

The quote is, I believe, for a vendor to bundle with their own software / service. They can source from the same place we do directly but are investigating if offloading the warranty (RTB / or supply and customer self install) and hassle of import / shipping etc is worth using someone like myself and paying a margin on top.

The brief is for upto 40 units over 2 racks. There is currently no information as to whether both racks will be in a single location but they are likely to be for a single end user.

but I do have an acecdote

Interesting, thanks for sharing. This is my thought too. I am likely to want to hedge against the servers becoming unavailable over the next 12 months by overstocking as if they do become unavailable I am stuck (give or take the two units I have for myself).

Just as a thought: each spare chassis at ~$250 is probably going to cost you less than even buying one motherboard separately so might be worth it to get an extra. As Andy and others have mentioned, for 12-24 months you should be relatively OK but in terms of labor just having a cold spare chassis around at each site may be the easiest thing to do.

Whilst logically very sound, rack availability may make that not possible. The site is not my own so I would most likely have to store the spares myself.

I feel that for the most part, these things don't fail. I have racks of compute clusters and the most often failed part is the HDD. I have one spare chassis (no memory or CPUs) with 4 sleds for the 10 that I have. Another group bout a similar spare shell for their 20 odd units. This should last us easily for the next 2-3 years, basically the useful life of the gear. I would not bother populating the spare parts.

The C6100s were produced by a separate unit in Dell from the standard and they have a pretty good reputation, it would seem. I would tend to agree on reliability but I will also need to factor in issues related to deracking, shipping to the supplier, shipping to me in Singapore. Whilst both suppliers I have dealt with have packaged well, working with a volume of up to 40 units, the probablility of hitting a problem is likely to rise a fair bit. The other off-shoot is a US$300 shipping round trip to send back a unit and get a replacement if it is not possible to just ship back a faulty part.

The economic model of used C6100 with their different position towards end of life and cheaper availability of whole systems influences the cost calculation.
1) It is reasonable to assume that used C6100's will continue to be available for the next 12-24 months, with potentially declining acquisition cost.

Possibly although Vista is reporting very few / no stocks, one of the suppliers for one of my own units is down to a few units left, the second supplier for my second unit seems to have stock but a couple of untried suppliers still report large availability. The issue would be one of confirming they will ship outside the US and the possibility that as the number of suppliers with stock lowers, whether the suppliers still 'in the game' will start to rise prices if demand is good.

2) The probabilty curve for warranty incidents has its peak at the beginning of a cycle

Fully agree. Neither of my units had an issue past a loose cable on one. Of course there is the work required to check and patch (bios / firmware) to get each unit up to spec.

3) Will you supply these servers only to one customer, or do you intend to broaden your C6100 business to multiple customers - spreading the probability further

One customer so a bit easier.

4) You might need less coverage by static things like case, vs. faster worn things like PSU, drives.

Yep so I may need to look at getting a couple of chassis plus a few more PSUs. Hard drives are not included thankfully . Ram can be sourced locally for a decent price and a couple of units spare should be able to cope with CPU failures if any, I would have thought.

5) You might overprovision on site, and agree with the customer a flexible warranty and/or payment model if they need to tap into the spare capacity on site. Reducing your labor cost in a warranty case.

Customers customer . Might not be an option but I am likely to offer a 'menu' of services as part fo the quote.

Take a look at google stat Google spotlights data center inner workings | News Blogs - CNET News and Google?s Disk Failure Experience

Thanks, I will have a good read through that link.

One of my own servers comes through Peter and he seems to have a fair amount of stock at this time so i will probably have a chat with him.

Other variable to through in is exchange rate fluctuations. Over the last year it has gone from 1.25 or 1.3 and back to 1.25 again. The difference on 40k of 0.05% plus international bank fees can make a fair difference. The rate is curently pretty low and fairly stable.

Thanks everyone for the insightful information. Lots to think about. Partly whether it is worth the risk .

RB