[Info-vax] OT: news from the trenches (re: Solaris)

Stephen Hoffman seaohveh at hoffmanlabs.invalid
Wed Mar 11 10:55:59 EDT 2015


On 2015-03-11 12:35:21 +0000, johnwallace4 at yahoo.co.uk said:

> [1] I see earlier this week HP announced a deal with Foxconn 
> for"Cloudline" servers which are said to appeal to Facebook/Google/etc 
> - basically a commodity low end server with none of the Proliant value 
> add, a box that has little consequence if it fails because no one 
> really cares about lost or corrupt data at FB or G, and you can throw 
> the box away when it fails because they're bought by the thousand on a 
> 2year lifecycle anyway. I'm not sure why that will now be an HP market 
> just by adding an HP badge and HP margins. More relevantly, massive but 
> lightweight scalability historically hasn't been the VMS market. It 
> probably still isn't, but my crystal ball is bust again.

Was looking for specs on those, though the Open Compute stuff is very 
interesting — standard server designs from various sources 
<http://blogs.microsoft.com/firehose/2014/01/27/microsoft-contributes-cloud-server-designs-to-the-open-compute-project/>, 
and with hardware available from various vendors, and where the buyer 
can also spec the associated external components.    Having standard 
connections and features means wider support, and not having to adapt 
to or to customize to the value-add features and protocols that are 
offered by specific server vendors.

As for reliability features, Intel has been moving what various folks 
have called RAS into their upper-end processors for some years.  
<http://labs.hoffmanlabs.com/node/95>   Intel are now selling RAS 
features on their lower-end Xeon Processor D SoC that they recently 
announced; RDIMM ECC, etc.

Yes, VSI is going to be dealing with what the Intel chips — or the AMD 
chips, if VSI decides to add support for specific features of those — 
are reporting up to the OS level for diagnostics or for host-OS 
assistance with a detected error or for logging purposes, with the 
recent chips.  What isn't automatically dealt with by the chips, that 
is.

Yes, the product lifetimes of the boxes are getting shorter.  
Conversely, the newer boxes can be more dense or more power-efficient — 
the older Itanium and Alpha boxes I work with are pretty hot, and 
they're slow.   As for reliability, if you use low-end low-cost on your 
memory or install the cheapest disks you can find — as can happen with 
an Integrity server or an AlphaServer, for that matter — you'll 
probably have memory errors or disk errors.  The bottom-end x86-64 
boxes that the value-conscious folks tend to buy are pretty awful, too. 
 The upper end boxes have correspondingly higher-grade parts, and 
better reliability.

As for what passes for VMS-related hardware reliability, DEC Field 
Service used to be into one data center I'm familiar with once or twice 
a month for random problems with any of five VAX boxes years ago — 
memory failures, failed disks, dead controllers, failed systems.   With 
newer gear, that frequency of maintenance is just not something that's 
required now, even with the value-conscious boxes.  Those VAX boxes 
cost US$50K for the middle configuration, and $100K for the upper-end 
configuration, per box.  I can afford to replace quite a few mid-upper 
ProLiant-grade x86-64 server boxes for that, and more often, and still 
have change left over for a nice steaming mug of tea.   That, and the 
newer boxes tend to be less expensive to power and cool, or I can stuff 
more boxes into less space in the data center.  Times change.


-- 
Pure Personal Opinion | HoffmanLabs LLC




More information about the Info-vax mailing list