Skip to content

Adding memory to a Sun Ray server

May 19, 2005

Before I went off to Italy our Solaris 10 Sun Ray server was struggling to cope with the 80+ users on it. This was entirely due to it only having 32Gb of memory. So I suggested that if we had the boards available that the system should get a memory upgrade over the weekend.

On checking when I returned the system now has 64Gb and is again performing perfectly, has picked still more users, and did not get rebooted so no user sessions were lost.

You have to love Dynamic Reconfiguration.

So thank you to the folks who did this.



From → Solaris

  1. Mike Gerdts permalink

    Dynamic Reconfiguration is cool. However…
    It seems as though this becomes a rather expensive approach compared to what you could get by balancing horizontally.
    Midrange or High-end cost The list price of a 32 GB 4×1.2GHz USIII board is 160k ($100k + after discount for pretty much all customers, I would guess). Add in to that the cost of the frame, slot 1 board(s), boot disks, network adapters, floor space, and support on the midrange or high end system, and the hardware starts to look kinda pricey compared to the cost of client-based computing.
    Entry level A V40z with 4 dual core Opterons and 32 GB of RAM has a list price of about $48k. This includes boot drives, IO, more powerful CPU’s, etc. Not only that, the hardware maintenance cost on a V40z is dirt cheap, even compared to a V240.
    Is your Sun Ray server a midrange/high-end server for reasons other than cost? That is, do you use the more expensive hardware because a SunRay farm really scales better vertically, or is this just a way for Sun to have more experience running 25k’s in house in production roles? Does the SunRay software provide a means to automatically direct new sessions at any of many (2 – 100) of a farm of SunRay servers?

  2. Our choice of server was slightly complex.
    We are an engineering department and we manage our server ourselves rather than having OPs do it. So this drives us to have as few servers as possible.When the decision was made Opeteron was not an option. We do now have an opteron server coming on line
    We also actively want to have a system that is different from those run by Ops so that we are testing more aspects of the system. If we are not doing something different from Ops then we would just use their servers.
    We then have 2 domains running the next release of solaris (as I have blogged about previously here:
    This allows us to make extensive use of DR and hence give it the best real life shake down we can so that when our customers need it we can be sure we have done everything we can to be sure it works.
    All these things factor in to the choice of a SunFire chassis. Add to that that we can then use these systems to keep warm the fastest boards that we have for the lab while they are not needed for escalations. If they are needed then we DR them out and replace then with the slower boards. Since we have to hold the stock any way this is a zero cost option (though I’m sure an accountant would tell me that there is some bizarre reason why it is not). It also gives us great confidence that the lab hardware is good.
    Yes the Sun Ray software can do load balancing across a number of servers. The OPs Sun Ray servers operate this way. Ours do not as we select the server based on the OS release

  3. William R. Walling permalink

    This ‘Sun Ray Server’ hardware discussion scares me.
    Clientele uptime is ‘not an option’.
    Any lost ‘session’ capability translates to clientele downtime within a ‘Sun Ray’ Thin Client environment.
    Does anyone at SUN view this as a ‘show stopper’?
    A ‘DCD’ solves this malady.
    I will try your recommended chainlube soon.

  4. Chris Saul permalink

    What’s a DCD?

  5. Chris Saul permalink

    What’s a DCD?

  6. William,
    I’m not sure I understand your point.
    In terms of session stability Sun Ray has a better story than a stand alone box since it becomes
    economical to use robust hardware with good RAS capabilities as the server as it can be shared by many users.
    If the Sun Ray fails (which IME is very rare) then I can plug a new one in and my session is still o.k.If the network fails the session is not lost either so when the network is fixed or if I move to another site where the network is still o.k. the session is still present.

  7. William R. Walling permalink

    Chris Saul,
    ‘DCD’ stands for ‘Digital Convergence Device’.
    A ‘DCD’ (appliance) works with ALL digital data within ANY environment.
    Thus far, Santa Clara is not interested.
    Chris Gerhard,
    ‘Failover’ is an expected ‘session’ protection option employed within a Sun Ray Server (hardware).
    However, ‘spinning’ mechanisms, of any type, continue to be championed and NOT eliminated.
    Server Hardware + Application Software + Transport Layer + Thin Client Hardware = Observed Latency.
    I invite a trial of 64 bit hardware with ‘true’ 64 bit code, view the results.
    (Hint: Tiger on a G5, 32 bit applications are readily identified as is 64 bit software.)
    Forthcoming ware from SUN will likely not solve this identified ‘latency’ problem.
    ‘DCD’ technology alleviates this observed issue.

  8. William,
    I’m sorry, but I just don’t understand what point you are trying to make.
    To implement session failover either requires that the session be stateless or the OS on which it runs provides a fault tolerant environment (which usually implies fault torlerant hardware) which is costly whether done in software or hardware.
    Given the high reliablity of modern systems particualy if they have DR the businness case for building a fault tolerant server is not particularly good as they tend to be slower than the non FT boxes and usually more than twice the price. That said I always thought there was a market for using the FT 1800 as a sunray server.
    As to latency, for the intended market there is no latency problem for Sun Ray. I sit here at my desk at home typing into a web browser running on a Server in the office and there is just no delay I can see. I’m sure there are apps that won’t work well here, games being one, but that is not the market that Sun Ray is addressing.

  9. William R. Walling permalink

    I am sorry for any confusion.
    Sun Ray addresses a ‘user’ market that global competition and non-profit organizations are now targeting.
    1) PPC will beget CELL for ‘all vertical markets’. Interesting, Mr. Torvalds now codes on PPC hardware.
    2) INTEL and AMD continue their tussle for ‘second place’ with x86 ware.
    3) CHINA leading ‘LC-PE-SFF’ global computing, Mr. Ruiz is about to receive a harsh ‘political’ lesson.
    A ‘PC’ replacement was just demonstrated at E3, it was not Redmond ware.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: